A mid-sized legal-tech vendor in Manchester recently audited where its company name surfaced in answers from ChatGPT, Perplexity and Google’s AI Overviews. The founder found that two niche legal technology indexes, one run by a bar association and the other by a procurement consortium, accounted for nearly every accurate citation, while three dozen general business listings produced none. The marketing director had spent the previous year submitting to anything that would accept a payment. So here is the question worth answering: when a company is trying to be cited, recommended or even just recognised by large language models, does the choice between an industry-specific index and a general one actually matter, and if so, by how much, and on what dimensions?
The directory visibility problem for AI
The discoverability problem has changed shape. For two decades, marketers spoke of search engine optimisation in terms of blue links: ten of them on a page, ranked by relevance, clicked by humans. Retrieval-augmented generation, conversational search, and AI-summarised answers have fragmented that model into something more diffuse. A company is no longer simply trying to rank; it is trying to be selected as a source by a language model that has been trained on a corpus, fine-tuned on additional data, and prompted to produce a synthesised answer from retrieved documents. Each of those three stages, pretraining, fine-tuning and retrieval, interacts differently with directory data, and each requires a different visibility logic.
Directories occupy a peculiar position in this new architecture. They aggregate structured data (names, categories, descriptions, URLs, contact details) in formats that are unusually friendly to machine ingestion. Crawlers from OpenAI, Anthropic, Google and Common Crawl pull directory pages because the schema is predictable and the content is dense with named entities. When a model needs to enumerate “vendors providing X service in Y region”, a well-structured directory entry is exactly the kind of evidence that retrieval systems prioritise. The catch: not all directories are weighted equally, and the weighting is opaque. Some are treated as authoritative; others are quietly downranked or filtered as spam.
The volume problem compounds the weighting problem. The number of indexes that accept business listings has expanded a great deal, partly through the proliferation of vertical SaaS marketplaces, partly through aggregators rebranding old data sets, partly through pay-to-play schemes. Listing breadth alone has diminishing returns; what matters is whether the listings appear in sources that AI systems actually treat as evidence rather than noise. The signal-to-noise ratio of any given directory determines whether inclusion produces citation lift or simply consumes time.
Then there is the trust dimension. Deloitte Insights reports that only 21% of enterprises surveyed have mature governance in place to manage the risks of agentic AI, a figure that has practical consequences for how AI systems handle source credibility. As models begin to incorporate stronger filtering for synthetic, low-quality, or commercially manipulated content, the directories most likely to keep their influence are those with editorial standards, verification processes and a clear domain focus. The directories most likely to be filtered out are those that accept any submission for a fee and produce thin pages that exist only to harbour outbound links.
The practical question for any business, whether a B2B SaaS firm, a regional services company, a manufacturer, or a clinical practice, is therefore not “should we be in directories?” but “in which directories does inclusion translate into measurable AI citation outcomes, and how do we choose?” The answer needs a comparative framework, because the existing heuristics (submit everywhere; submit only to your industry; submit only where your competitors are) each fail in characteristic ways. The rest of this article sets out a named framework, SIGNAL, that practitioners can apply to make the comparison rigorous.
Why current listing strategies fail
The generalist spray approach
The first failure mode is familiar to anyone who has inherited a marketing function from a predecessor with a credit card and an afternoon to spare. The approach: identify every directory that will accept a submission, fill in the form, pay the fee where required, repeat. The logic is volumetric, where more listings equals more backlinks equals more visibility equals more citations. That logic was never quite right for traditional SEO; it is decisively wrong for AI visibility.
The first problem is dilution. When a company appears in three hundred directories of varying quality, retrieval systems trained to weight by source authority encounter a long tail of low-quality listings that drag down the aggregate signal. A model presented with the question “what are the leading providers of supply chain visibility software?” will not enumerate every vendor in every directory it has ever crawled; it will sample from sources that its training and retrieval logic treat as authoritative. A spray of low-authority listings does not contribute to that sampling.
The second problem is consistency drift. Spray submissions are typically delegated to junior staff or outsourced to virtual assistants, and the descriptive copy varies from listing to listing, sometimes deliberately, to avoid duplicate-content penalties that no longer meaningfully exist, sometimes accidentally, because nobody maintained a canonical version. AI systems performing entity resolution across thousands of listings encounter conflicting descriptions and respond by either picking one arbitrarily or hedging the entity description into vagueness. Large-scale entity resolution work in scholarly publishing (Wiley reports that over 50% of its Online Library traffic comes directly from search engines like Google and Google Scholar) suggests that consistent metadata is a precondition for being surfaced reliably; the inverse is also true.
The third problem is opportunity cost. Time spent submitting to a third-rate aggregator is time not spent securing inclusion in a first-rate vertical index, building a relationship with a trade body that publishes member listings, or producing the editorial content that primary directories tend to require. The spray approach optimises for activity metrics (submissions completed) rather than outcome metrics (citations generated). A team measured on the former will produce more of the former and less of the latter.
The fourth problem, and the one that practitioners discover too late, is reputational. Some general aggregators have themselves been flagged by major search engines and language model training pipelines as low-quality sources. Inclusion in those aggregators is not neutral; it can mark an entity as participating in low-quality link networks. When models perform source-quality filtering, the entity’s other listings may be evaluated through that lens. The spray approach sometimes produces negative visibility lift.
The niche-only blind spot
The opposite failure mode is the purist position: list only in industry-specific directories, on the theory that vertical relevance is the only signal that matters. This is closer to correct than the spray approach, but it has its own problems, and they are less obvious because they show up as absence rather than presence.
The first problem is that AI systems do not retrieve only from vertical sources. When a user asks Perplexity for “good fintech vendors in London”, the model may retrieve from a fintech-specific index, but it may also retrieve from a regional business compendium, a chamber of commerce listing, a general technology marketplace, and a journalist’s published list. A company that has secured inclusion only in vertical sources is invisible to half the retrieval paths that lead to its category.
The second problem is the cross-vertical query. Many real-world questions cross category boundaries. “Vendors providing AI-powered supply chain analytics for the pharmaceutical sector” intersects three verticals, and the retrieval systems that answer such queries draw on sources that span verticals. A vertical-only listing strategy is poorly positioned for these intersecting queries because it lacks presence in the broader sources that join verticals together.
The third problem is category drift. The categories used by industry-specific directories were typically defined years before the current generation of AI products existed. A company building, say, autonomous procurement agents may not fit neatly into any existing category in a procurement directory; it may be classified under “e-procurement software” or “spend analytics”, neither of which captures what the product does. Vertical directories that have not updated their taxonomy misclassify emerging entities, and AI systems retrieving from those directories inherit the misclassification.
The fourth problem is regional coverage. Industry-specific directories are often anchored in a particular geography (a US-based fintech index, a German engineering directory, a UK legal services compendium), and a vertical-only strategy that selects the canonical directory for each market still leaves gaps where no canonical directory exists. The World Bank’s industrial policy work documents that low-income economies target growth in thirteen industries on average, more than twice the number high-income economies do. The implication for directory strategy is that the granularity of vertical categorisation varies dramatically by region, and a uniform niche-only approach will under-serve markets where vertical infrastructure is sparse.
The fifth problem is the credibility ceiling. Vertical directories tend to be smaller, less frequently updated, and less frequently crawled than the largest general indexes. They may be the most authoritative sources for their domain, but they are not always the most-visited. A purely vertical strategy concentrates risk: if the one or two leading vertical directories in a sector are slow to crawl, technically poor, or deindexed for any reason, the company has no fallback presence in larger indexes that might compensate.
Introducing the SIGNAL directory framework
SIGNAL is a five-component scoring framework for evaluating any directory’s contribution to AI visibility. The acronym expands to: Specificity of audience match, Indexing frequency by AI crawlers, Genuine authority signals, Niche citation density, and Linkage to knowledge graphs. Each component is scored on a 0 to 10 scale; the aggregate score determines whether a given directory belongs in a company’s listing portfolio, and at what priority level. The framework is deliberately neutral on the industry-specific versus general question. It is designed to surface directories of either type that perform well, and to filter out directories of either type that do not.
Before walking through each component, two design decisions deserve explanation. First, the framework weights audience match and authority more heavily than raw indexing frequency, on the assumption that being cited by a model in response to relevant queries matters more than being indexed in absolute terms. Second, the framework treats knowledge-graph linkage as a distinct dimension from authority, because a directory can be authoritative without being well-connected to the structured data layer that retrieval systems increasingly use to disambiguate entities. The two correlate but are not identical.
Specificity of audience match
The first component asks: how closely does the directory’s audience match the audience the listed business is trying to reach? Specificity is not the same as narrowness; a broad directory can have a highly specific audience if its users are predominantly the buyers, evaluators or referrers the business cares about. A vertical directory can have a poor audience match if its users are predominantly competitors, students or vendors selling into the listed companies rather than buying from them.
To score this dimension, three sub-questions help: who actually visits the directory; who actually queries AI systems in ways that surface the directory; and what proportion of either group represents commercial intent relative to the listed business. Traffic analytics tools that report referrer breakdowns can answer the first; AI citation tracking tools that record when a directory appears as a source can answer the second; the third typically requires qualitative judgment based on the directory’s positioning and editorial focus.
A directory scoring 9 or 10 on specificity is one whose audience is dominated by the exact buyer persona the business serves, with minimal noise from adjacent or irrelevant audiences. A directory scoring 4 or 5 is one whose audience overlaps partially with the target persona but is diluted by other groups. A directory scoring 1 or 2 is one whose audience is largely irrelevant to the business’s commercial goals, even if the directory is otherwise well-trafficked.
Worked example: a company selling clinical decision support software to hospital procurement teams might score a directory maintained by a hospital purchasing consortium at 9 (audience is procurement decision-makers in target accounts), a general healthcare technology marketplace at 6 (audience includes the target persona but also many adjacent personas), and a directory of medical device manufacturers at 3 (audience is mostly other vendors, not buyers).
Indexing frequency by AI crawlers
The second component measures how often AI training and retrieval systems actually crawl the directory. A directory whose pages have not been refreshed in the index of major retrieval systems for six months is contributing little to AI visibility, whatever its other merits. Indexing frequency is partly a function of the directory’s technical health (robots.txt configuration, sitemap submission, server response times) and partly a function of the directory’s perceived authority among crawlers, which determines how aggressively they revisit it.
Practical scoring means checking whether the directory’s pages appear in the cache of major search engines, whether the directory itself is referenced in AI system citations, and whether the directory shows recent crawler activity in its own analytics (a question for the directory operator, not always answerable externally). Tools that monitor AI citation patterns, several of which have emerged in the last eighteen months, can indicate whether a directory’s listings are being retrieved by which models.
A directory scoring 9 or 10 is one whose pages are crawled weekly or more frequently by major AI systems and whose listings appear regularly in AI-generated answers. A directory scoring 5 or 6 is crawled monthly and appears occasionally as a source. A directory scoring 1 or 2 is rarely crawled and seldom surfaces as a citation source, which suggests either technical problems or a credibility deficit on the crawler side.
Indexing frequency interacts with the agentic AI governance gap noted earlier. Deloitte’s finding that only 21% of enterprises have mature governance for AI risk management has implications here too, because directories without clear editorial standards are increasingly likely to be deprioritised by retrieval systems applying source-quality filters. A directory with high crawl frequency but declining citation appearance may be experiencing exactly that kind of filtering.
Genuine authority signals
The third component evaluates whether the directory carries genuine authority signals, the markers that retrieval systems and search algorithms use to estimate trustworthiness. These include the directory’s age and operational history; its editorial standards and verification processes; its institutional affiliations (a trade body, a university, a recognised media brand); its citation patterns in scholarly or trade publications; and the consistency of its data quality over time.
Authority is the dimension most easily faked by directory operators and most often misjudged by submitters. A directory may have high domain authority scores in tools like Moz or Ahrefs while having no editorial process whatsoever, because those tools measure link patterns rather than editorial integrity. A directory with modest external link metrics may carry substantial authority because it is operated by a respected institution and used by serious buyers in its sector.
To score genuine authority, the analyst should ask: does the directory verify submissions? Does it have editorial guidelines that are visibly enforced? Is it cited by name in trade press or academic literature? Is its operating organisation recognisable and reputable? Does it have a track record of removing fraudulent or low-quality listings? A directory scoring 9 or 10 answers yes to all of these. A directory scoring 5 or 6 has mixed signals, perhaps editorial guidelines but inconsistent enforcement. A directory scoring 1 or 2 has no apparent editorial process and accepts any submission for a fee.
The IEEE offers an instructive case in genuine authority signals. Its Impact Creators programme, described on the IEEE public visibility page, focuses on the stories of the people behind innovations rather than the technology itself, a positioning that signals editorial intent rather than mere aggregation. Directories that articulate a comparable editorial philosophy, even if smaller, tend to score higher on this dimension than larger directories that simply collect entries.
Niche citation density
The fourth component, niche citation density, measures the concentration of relevant entities in the directory and the patterns by which they reference one another. This is distinct from raw size: a directory with ten thousand listings spanning every industry has lower niche citation density for any given vertical than a directory with five hundred listings concentrated in one vertical. Density matters because retrieval systems looking for “the leading providers of X” are more likely to surface results from sources where many providers of X coexist, providing the comparative context that generative answers need.
Density also matters because internal linking within a directory, where entries reference one another via “related vendors”, “competitors” or “alternatives”, creates the kind of structured comparative data that AI systems can exploit. A directory with rich internal linking effectively provides a small graph of the sector, which is more useful to a retrieval system than an unconnected list.
To score niche citation density, the analyst should examine the directory’s coverage of the relevant vertical (what proportion of the recognised players are listed?), the depth of each listing (does it include comparative attributes, customer references, integration partners?), and the internal linking structure (are entries cross-referenced in meaningful ways?). A directory scoring 9 or 10 has comprehensive coverage of the vertical with rich internal linking. A directory scoring 5 or 6 has partial coverage with some linking. A directory scoring 1 or 2 has sparse coverage and isolated listings.
Worked example: in B2B procurement software, a directory that lists 80% of the recognised vendors in the category, includes detailed attribute tags for each, and cross-references them by integration partner would score 9. A general business directory that lists three procurement vendors out of dozens, without tags or cross-references, would score 2, even if its overall traffic dwarfs the vertical directory.
Linkage to knowledge graphs
The fifth component evaluates how well the directory connects to structured knowledge graphs (Wikidata, Crunchbase, Google’s Knowledge Graph, industry-specific ontologies) that AI systems use for entity disambiguation and enrichment. A directory whose entries map cleanly to canonical entity identifiers in Wikidata, for example, contributes data that retrieval systems can connect to other sources about the same entity. A directory whose entries float as standalone strings cannot be cross-referenced as easily.
Linkage is signalled by several technical features: structured data markup (schema.org Organization, LocalBusiness, Service), explicit references to Wikidata or other canonical IDs, consistent entity naming aligned with established conventions, and outbound links to authoritative sources about each listed entity. Directories that participate in linked open data initiatives or publish their data sets in machine-readable formats score particularly well on this dimension.
To score linkage, the analyst should inspect a sample of directory pages for structured data, check whether listed entities have Wikidata equivalents that are explicitly referenced, and assess whether the directory provides the kind of identifiers that retrieval systems can use for disambiguation. A directory scoring 9 or 10 has comprehensive structured data and explicit knowledge graph linkages. A directory scoring 5 or 6 has partial structured data without graph references. A directory scoring 1 or 2 has unstructured listings with no machine-readable identifiers.
This dimension has become more important as AI systems increasingly rely on knowledge graphs to ground their answers. Research published in a published examination of structured data adoption among business directories suggests that linked, schema-marked entries are retrieved by AI systems at substantially higher rates than equivalent unstructured listings, even when the underlying content is identical. The scoring of this dimension reflects that asymmetry.
Applying SIGNAL to directory selection
Scoring industry-specific directories
Industry-specific directories, by their nature, tend to score well on specificity and niche citation density and less consistently on indexing frequency, authority and knowledge-graph linkage. The reason is structural: vertical directories are usually built and maintained by people deeply embedded in the sector, who have strong intuitions about audience and category coverage but may lack the resources or technical sophistication to maintain crawl-friendly infrastructure or schema markup at the level of the largest general indexes.
A typical scoring exercise for a well-regarded vertical directory might produce: specificity 9 (audience is precisely the target buyer persona); indexing frequency 5 (crawled monthly, with some technical limitations); authority 8 (operated by a recognised trade body with editorial standards); niche citation density 9 (comprehensive vertical coverage with internal linking); knowledge graph linkage 4 (basic schema markup, no Wikidata references). Aggregate: 35 out of 50, or 70%.
The exercise shows where the vertical directory’s strength concentrates and where its gaps lie. A 70% directory is worth listing in, but a sophisticated visibility programme would also work with the directory operator, where possible, to improve the dimensions that score lower. Some operators welcome such conversations; others do not. Whether an operator will engage on technical improvements is itself a signal of long-term value.
A weaker vertical directory might score: specificity 8 (still vertical-focused but with broader audience including students and competitors); indexing frequency 3 (irregular crawls, technical issues); authority 4 (commercial operator with limited editorial process); niche citation density 5 (partial coverage); linkage 2 (no structured data). Aggregate: 22 out of 50, or 44%. This directory is borderline, worth listing in only if the cost is minimal and the alternative cost of not appearing in any vertical source is high.
Scoring general directories
General directories present the inverse profile. They tend to score well on indexing frequency, authority and sometimes knowledge-graph linkage, because they are run by larger operations with more technical resources, and less consistently on specificity and niche citation density. The largest general indexes have the budgets to maintain modern infrastructure, implement schema markup at scale, and take part in linked-data initiatives; what they cannot easily provide is the audience concentration of a vertical source.
A typical scoring exercise for a strong general directory might produce: specificity 5 (audience overlaps with target but is diluted by many other personas); indexing frequency 9 (crawled weekly, surfaces in AI citations regularly); authority 8 (long-established operator with editorial standards); niche citation density 4 (some vertical coverage but limited internal linking within the vertical); linkage 8 (comprehensive structured data, knowledge graph references). Aggregate: 34 out of 50, or 68%.
The aggregate score is similar to the strong vertical directory above, a finding that recurs in the application of SIGNAL across many sectors. Strong general and strong vertical directories often produce comparable aggregate scores via different paths, which is one of the central insights the framework yields. Neither type dominates the other in the abstract; the optimal portfolio is determined by which specific directories of each type are available in the relevant vertical, and how their dimensional profiles complement one another.
A weaker general directory, the sort that proliferated in the 2010s and now makes up much of the directory landscape, might score: specificity 2 (audience is largely irrelevant to any specific buyer persona); indexing frequency 5 (still crawled, but trust-filtered); authority 3 (no meaningful editorial process); niche citation density 2 (sparse vertical coverage); linkage 4 (basic markup). Aggregate: 16 out of 50, or 32%. This directory is not worth listing in; the time cost exceeds any plausible visibility return.
Weighting by business vertical
The aggregate scores produced by SIGNAL are useful, but the dimensional weights should not be uniform across all business verticals. Different verticals have different visibility dynamics, and the relative importance of the five dimensions varies accordingly. A few illustrative weightings clarify the point.
For B2B SaaS targeting enterprise buyers, specificity and niche citation density tend to dominate, because enterprise procurement processes lean heavily on vendor comparison within tightly defined categories. A weighting of 1.5x on specificity and 1.3x on niche citation density, with 1.0x on the others, captures this reality. Under such weighting, vertical directories often outperform general ones on the weighted aggregate.
For local services targeting consumer buyers, authority and indexing frequency become more important, because consumer queries often reach AI systems through general-purpose interfaces that draw heavily on widely-trusted, frequently-crawled sources. A weighting of 1.4x on authority and 1.3x on indexing frequency tilts the balance toward strong general directories with local coverage, while vertical directories serve a complementary role.
For regulated industries such as financial services, healthcare and legal, authority and knowledge-graph linkage become disproportionately important, because compliance-sensitive AI deployments increasingly filter sources by institutional credibility and structured-data verification. A weighting of 1.5x on authority and 1.4x on linkage produces aggregate scores that favour directories operated by professional bodies, regulators, or recognised academic institutions.
For emerging technology categories, niche citation density and specificity become critical, because the categories themselves are still being defined, and the directories that articulate clear taxonomies effectively shape how AI systems describe the category. Early inclusion in the directory that becomes the canonical reference for an emerging category produces visibility returns that compound for years.
Setting minimum threshold scores
Not every directory that scores above zero is worth listing in. Practical implementation of SIGNAL needs minimum threshold scores below which directories are excluded from the portfolio regardless of how cheap or easy inclusion would be. Two thresholds matter: the absolute minimum (below which inclusion is actively counterproductive) and the priority threshold (above which inclusion is worth significant effort).
Empirically, an absolute minimum aggregate score of around 25 out of 50 (50%) tends to separate directories worth considering from those worth avoiding. Below 50%, the directory is likely to provide no measurable AI visibility lift and may carry minor reputational risk through association with low-quality networks. The exact threshold varies by vertical and by the company’s risk tolerance, but 50% is a reasonable default.
The priority threshold is higher, typically around 35 out of 50 (70%). Directories above this threshold deserve dedicated effort: high-quality submissions, ongoing maintenance, relationship-building with operators where possible, and inclusion in regular audit cycles. Directories between 50% and 70% deserve inclusion but lower priority, submitted once, maintained quarterly, not invested in beyond that.
Threshold-based filtering produces a more focused portfolio than unfiltered submission. A typical company applying SIGNAL with thresholds finds that of the hundred or so directories it might theoretically list in, perhaps fifteen to twenty-five clear the priority threshold, another ten to twenty clear the minimum threshold, and the remainder are excluded. The discipline of exclusion is as important as the selection of inclusions; it is what frees capacity for sustained engagement with the directories that actually produce returns.
Building a hybrid listing portfolio
Once thresholds are applied, the resulting portfolio almost always includes both industry-specific and general directories, which is the empirical answer to the question that titles this article. Neither type alone produces optimal coverage; the combination does. The art lies in the mix and in the role each directory plays.
A representative hybrid portfolio for a mid-sized B2B company might include: three to five top-tier vertical directories (high specificity, high niche citation density); two to three top-tier general directories (high indexing frequency, high authority, broad reach); two to three regional directories where the company operates (geographic specificity, local authority); one or two adjacent-vertical directories (capturing cross-vertical queries); and selective inclusion in two or three structured-data-heavy directories that prioritise knowledge graph linkage. In total, roughly twelve to sixteen directories, each chosen because its dimensional profile contributes something the others do not.
The portfolio approach mirrors investment thinking: diversification across complementary risk profiles. No single directory is treated as decisive; the aggregate produces resilience against any individual directory’s decline or deindexing. The Deloitte global industries page describes how Deloitte itself organises around six industries and twenty sectors, a structure that reflects the recognition that industry coverage needs both breadth and depth, with each sector providing specificity and the industry framework providing comparability across sectors. The same logic applies to directory portfolios: breadth across types provides robustness, depth within each type provides specificity.
Maintenance burden is the practical constraint on portfolio size. Each directory in the portfolio needs periodic verification that the listing is current, accurate and consistent with the company’s canonical description. A portfolio of fifteen high-quality directories represents perhaps two to four hours of maintenance per quarter; a portfolio of fifty mediocre directories represents twenty hours per quarter and produces lower aggregate visibility. The maintenance ratio is a useful sanity check on portfolio composition.
Tracking AI citation outcomes
SIGNAL is a selection framework; it must be paired with an outcome tracking framework to confirm that the selected directories are producing the expected visibility lift. The outcome metrics that matter are: AI citation frequency (how often does the company appear as a source in AI-generated answers to relevant queries?); AI citation accuracy (when cited, is the company’s information correctly represented?); referral patterns (when AI citations include links, do they convert to traffic?); and competitive share-of-voice (in answers that enumerate vendors, where does the company rank?).
Several tools have emerged to track these metrics, among them Profound, Otterly, Athena and Goodie, though the category is young and the methodologies vary. The common approach involves running representative queries against major AI systems on a regular cadence and recording which sources are cited. Cross-referencing the cited sources against the directory portfolio reveals which directories are actually contributing to citations and which are passive.
The outcome data should feed back into SIGNAL scoring. A directory that scored highly on the framework but produces no observable citation lift over a six-month period should be reviewed and possibly downgraded; the dimensional scoring may have missed something the actual retrieval behaviour reveals. A directory that scored moderately but consistently appears as a citation source should be upgraded and given more attention. The framework is designed to be iterative; static scoring is less valuable than dynamic scoring informed by observed outcomes.
Citation tracking also reveals the directories that AI systems treat as authoritative for the company’s specific category, which can differ from the directories the company itself selected. Sometimes the unsought citation source is a directory the company had not considered listing in; that is a signal to investigate inclusion. Sometimes the unsought citation source is a competitor’s customer list or case study repository; that suggests a different visibility tactic. The tracking discipline produces information that no a priori framework can fully predict.
Worked example: a B2B SaaS company
Consider a mid-sized B2B SaaS company, call it FlowMetrics, selling supply chain visibility software to mid-market manufacturers. The company has $25M ARR, sells primarily in North America and Europe, and is trying to increase its presence in AI-generated answers to queries like “supply chain visibility software for manufacturers” and “alternatives to [larger competitor]”. The marketing director has inherited a list of forty-three directories the company is currently listed in, with no clear logic behind the selection.
The first step is auditing the existing portfolio against SIGNAL. The team scores each of the forty-three directories on the five dimensions, producing an aggregate score and a 50%/70% threshold classification. The audit reveals six directories above 70%, eleven between 50% and 70%, and twenty-six below 50%. The twenty-six below threshold include several pay-to-play general directories, a half-dozen geographic directories the company has no presence in, and several aggregators of low quality. These are flagged for delisting where possible (some directories make delisting difficult) and for explicit deprioritisation in any case.
The six above-70% directories include: the primary supply chain technology vertical directory (specificity 9, indexing 7, authority 8, niche citation density 9, linkage 6 = 39); a manufacturing technology marketplace operated by a recognised trade body (8, 7, 8, 8, 7 = 38); a general B2B software comparison platform with strong supply chain coverage (5, 9, 8, 6, 9 = 37); a Gartner-aligned vendor matrix listing (7, 7, 9, 7, 7 = 37); a regional manufacturing directory in the company’s primary North American market (8, 6, 7, 8, 6 = 35); and a structured-data-heavy general business directory with manufacturing schema (4, 9, 7, 5, 10 = 35). The mix is exactly the hybrid pattern SIGNAL tends to produce: vertical directories scoring well on specificity and density, general directories scoring well on indexing and linkage, with comparable aggregate scores.
The eleven between-thresholds directories include several adjacent-vertical sources (logistics, ERP, manufacturing operations) that capture cross-vertical queries; two regional directories in European markets where FlowMetrics has minor presence; and three general directories with moderate authority and indexing. These are kept in the portfolio at lower priority, submitted once, maintained quarterly.
The next step is identifying gaps. The team conducts a citation-tracking exercise across major AI systems for a basket of relevant queries and records which sources appear as citations. Two findings emerge. First, four of the six top-priority directories appear regularly as citations (the two that do not, the regional manufacturing directory and the structured-data-heavy general directory, appear less often than expected). Second, three directories that are not in FlowMetrics’s portfolio appear repeatedly as citation sources for relevant queries. These are flagged for evaluation and submission.
The two underperforming top-priority directories are investigated. The regional manufacturing directory turns out to have a technical issue: its company pages are not being included in its sitemap, and most AI crawlers are reaching only the category pages, not the individual entries. The team contacts the operator and supplies the necessary technical detail; within six weeks, the directory begins appearing as a citation source. The structured-data-heavy general directory turns out to be cited mostly for queries about much larger companies; FlowMetrics’s listing is technically correct but not in the company-size range that the directory’s audience tends to query about. The team accepts the limitation and downgrades the directory to medium priority.
The three new directories identified through citation tracking turn out to be: an academic-industry consortium directory at a major university (high authority, decent specificity); a journalist-curated list of supply chain technology vendors maintained by a trade publication (high authority, very high specificity); and a procurement-focused vendor index used by Fortune 500 procurement teams (moderate indexing, very high specificity for a critical buyer persona). All three are pursued; two accept submissions immediately, one requires editorial review and several months of relationship-building before inclusion.
Six months after the audit, FlowMetrics’s citation frequency for its target queries has roughly doubled, share-of-voice in vendor enumeration queries has improved from sixth most-mentioned to third, and traffic from AI referral sources has tripled from a low base. The team attributes most of the improvement to three factors: the inclusion in the three previously-unidentified directories; the technical fix on the regional manufacturing directory; and the consistent canonical description across all priority listings (the audit also discovered that descriptions had drifted across listings, and harmonising them produced measurable improvement). The improvement is not equally distributed across directories, with three directories accounting for roughly 60% of the citation lift, which is consistent with the power-law dynamics that tend to characterise this domain. a 2024 review the typical pattern where a small subset of a portfolio drives most of the visibility outcomes, even when all members of the portfolio meet quality thresholds.
The lesson FlowMetrics drew from the exercise: the audit and selection work was the easy part; the maintenance, technical liaison and ongoing monitoring were the hard part, and the source of most of the actual lift. Without a maintenance discipline, the audit’s gains would have decayed within a year as listings drifted, contact details changed, and operators shifted their own priorities.
Worked example: a local healthcare practice
The second worked example is a regional healthcare practice, call it Westbridge Orthopaedics, operating four clinics in the south of England. The practice serves both NHS-referred patients and a growing private patient base; the marketing question is how to increase visibility in AI-generated answers to queries like “best orthopaedic surgeons near [city]”, “knee replacement specialists in [region]”, and “private orthopaedic clinics [postcode]”. The practice manager has historically focused on Google Business Profile and a handful of healthcare-specific directories but suspects the strategy is not optimised for AI visibility.
The audit covers nineteen existing directory presences, ranging from the obvious (Google Business Profile, NHS Choices, Doctify, Top Doctors) to the marginal (general regional business directories, miscellaneous healthcare aggregators). SIGNAL scoring produces five directories above 70%, six between 50% and 70%, and eight below.
The above-70% directories illustrate the regulated-industry weighting tilt described earlier. Authority and knowledge-graph linkage carry premium weight in healthcare, where AI systems are increasingly cautious about citing sources for medical queries. The five top directories are: NHS Choices (specificity 9, indexing 9, authority 10, niche citation density 8, linkage 8 = 44); Doctify (8, 8, 8, 7, 7 = 38); Top Doctors (7, 7, 7, 8, 7 = 36); the General Medical Council’s specialist register (5, 7, 10, 5, 9 = 36, with authority weighted heavily); Google Business Profile (6, 9, 9, 4, 10 = 38). The mix is again hybrid, with vertical health-specific sources alongside the general but high-authority Google Business Profile.
The mid-priority directories include private medical insurer-aligned listings (BUPA-recognised provider lists, AXA PPP-recognised lists), a regional NHS trust referral directory, two academic-institution-affiliated specialist registers, and a couple of patient-review aggregators. The below-threshold directories include several general regional business indexes, a couple of healthcare aggregators with poor editorial standards, and a directory the practice was paying GBP 400 a year to be listed in but which produced no observable visibility lift.
The citation-tracking exercise produces several useful findings. AI systems queried about orthopaedic services in the relevant region cite NHS Choices and Doctify most frequently, which is predictable, but they also cite, more often than expected, two sources the practice had not considered: a regional private healthcare guide published by a local newspaper group, and a patient-experience platform operated by a major hospital network. Both are evaluated and pursued.
The exercise also reveals an authority gap. AI systems hedge significantly when answering medical queries, often refusing to make recommendations or specifying that users should verify with regulated sources. The practice’s listings on commercial directories, while accurate, do not carry enough authority weight to overcome the hedging. The practice manager identifies this as a structural limitation rather than a tactical one; the response is to invest in authority-building activities (research collaborations, GMC-recognised specialist accreditations, peer-reviewed publications by clinicians) that feed back into directory listings as upgraded credentials. Over a twelve-month period, this slow-burn authority investment produces more citation lift than any single directory submission, because it improves the company’s profile across many directories at once.
According to this case study of healthcare visibility patterns, regulated-industry visibility outcomes are dominated by authority signals to a degree that exceeds most other verticals; the worked example confirms the pattern. The practice’s eventual portfolio settles at about ten high-quality directory presences, with disproportionate investment in the three or four that carry institutional authority. The remaining listings are maintained but not actively cultivated.
Two limitations are worth noting from this example. First, healthcare visibility is constrained by regulatory factors that limit how aggressively any practice can market itself, particularly through AI channels where the boundary between recommendation and information is contested. Second, the AI systems’ caution about medical queries means that visibility lift is harder to achieve and measure than in commercial verticals; the practice manager had to develop more sophisticated tracking to capture lift signals that were smaller in absolute terms but meaningful in context.
Edge cases and framework limitations
Emerging industries without directories
The first edge case is the company operating in an industry too new to have established directories. Consider a startup building autonomous AI agents for legal contract review. The category, call it “agentic legal AI”, did not exist three years ago, and no canonical directory for it has yet emerged. Traditional legal-tech directories include the company under “contract review” or “legal AI”, but those categories conflate it with quite different products. General AI directories list it under categories that conflate it with consumer chatbots. Vertical legal directories are not technical enough to evaluate it correctly.
SIGNAL applied to this situation produces low scores across the board, because no directory matches the audience precisely or covers the niche densely. The framework correctly identifies that the directory landscape is failing the company, but it does not, by itself, prescribe a solution. The solution typically combines four tactics: advocate within the most relevant existing directory for a new category that better captures the product; accept partial-fit listings under the closest available categories while pushing for taxonomic updates; invest in primary content (documentation, case studies, technical comparisons) that can serve as a citation source even in the absence of directory infrastructure; and work with industry analysts and trade press whose coverage may itself function as a quasi-directory.
The fourth tactic deserves elaboration. Industry analyst reports, trade press category overviews, and journalist-curated lists increasingly serve as the de facto directories for emerging categories. AI systems trained on broad web crawls treat these sources as authoritative for category-defining queries, often more than commercial directories. A startup in an emerging category may achieve more visibility lift from inclusion in three carefully-chosen analyst reports than from inclusion in twenty marginal directories. SIGNAL can be extended to score these quasi-directory sources on the same dimensions, with adjustments for their different structural characteristics.
The framework limitation here is that SIGNAL assumes a directory landscape exists. When it does not, the framework can diagnose the absence but cannot fully substitute for the missing infrastructure. A company in an emerging category should expect to invest in creating directory-like resources (community indexes, open-source comparison repositories, technical benchmarks) that may eventually become the canonical sources for the category. The investment is long-term but compounding; companies that successfully establish such resources gain durable visibility advantages over later entrants who must accept the established taxonomy.
Cross-vertical service providers
The second edge case is the service provider whose work spans multiple verticals. Consider a management consulting firm specialising in AI strategy, serving clients in financial services, healthcare, manufacturing, and retail. SIGNAL scoring for this firm is complicated by the fact that the firm should plausibly score directories well in all four verticals, but a single firm cannot maintain a top-priority presence in all four vertical landscapes; the maintenance burden becomes prohibitive.
The resolution requires explicit prioritisation by revenue concentration. If 60% of the firm’s revenue comes from financial services, the financial services vertical directories receive top-priority treatment, and the other vertical directories receive medium-priority treatment. As revenue mix shifts, priorities shift accordingly. The framework is not violated by this approach, but it needs an additional layer of business-context weighting beyond the dimensional scoring.
An additional complication is that AI systems answering cross-vertical queries such as “AI strategy consulting for healthcare” draw on the intersection of multiple sources, and a firm that is well-listed in healthcare AI directories will surface for the cross-vertical query even if it is not listed in horizontal management consulting directories. The implication: for cross-vertical service providers, the optimal portfolio is often deeper in fewer verticals rather than shallower across all of them. The horizontal listings serve as backup; the vertical listings do the work.
A second-order complication concerns canonical description. A firm serving four verticals must decide whether to use one canonical description across all listings or to tailor descriptions to each vertical. The evidence on this question is mixed. Tailored descriptions tend to perform better in vertical-specific queries; canonical descriptions tend to produce more consistent entity resolution across the broader web. The pragmatic compromise is a stable canonical “spine” (name, founding date, headquarters, leadership) combined with vertical-specific service descriptions that vary by listing. The spine ensures entity coherence; the variable elements optimise for vertical relevance.
Paid vs free directory tradeoffs
The third edge case concerns the tradeoff between paid and free directory inclusion. SIGNAL scores quality without explicit reference to cost; in practice, cost matters because budgets are finite and some high-scoring directories charge significant fees. The question is whether paid inclusion in a high-scoring directory outperforms free inclusion in a slightly lower-scoring one, when the budget would otherwise support inclusion in both.
The empirical answer depends on the specific directories, but several general observations hold. First, the relationship between paid status and visibility lift is non-linear; some paid features (enhanced listings, sponsored placements) produce measurable lift, others (logo display, premium badges) produce mostly aesthetic returns that do not translate into AI citations. Before paying for premium features, a company should request data on what those features actually change in the listing’s structure and presentation; if the change is purely visual, AI systems will not see it.
Second, some paid directories restrict free listings in ways that affect SIGNAL scores. A directory that allows free entries but limits them to bare-bones fields effectively reduces the niche citation density of the free entry. In such cases, the paid upgrade may be the difference between a low-density and a high-density listing, which can substantively shift the SIGNAL score and the resulting visibility lift.
Third, paid inclusion in low-quality directories is almost always negative-value. The fact that a directory charges for inclusion does not make it authoritative; many low-quality directories charge specifically because they cannot attract submissions on merit. Pay-to-play directories with no editorial process should be excluded regardless of price; the cost is not the issue, the directory is.
Fourth, the highest-authority directories often have non-monetary inclusion criteria that are more demanding than any payment: editorial review, institutional affiliation, peer endorsement. Securing inclusion in these requires investment of a different kind: time, relationships, content, credentials. A company comparing paid versus free options should also weigh the more demanding inclusion paths that produce the most durable visibility, even when they do not appear on a budget line. According to a study available here, the directories that demand editorial review tend to produce citation outcomes that compound over time at rates several multiples of those produced by simple paid inclusion, although the upfront investment is significantly higher; the ROI calculation must take the time horizon into account.
Implementing SIGNAL in your workflow
Auditing existing directory presence
Implementing SIGNAL begins with an audit of existing directory presence, typically more extensive than the team expects. Most companies discover, on auditing, that they appear in directories they did not knowingly submit to, that descriptions vary across listings in ways nobody managed, and that some listings reference outdated personnel, addresses or product names. Before applying the framework, the audit’s first task is to inventory what exists.
The inventory process combines several techniques. A search for the company name across the web reveals most listings, though it misses some directory pages that are not well-indexed. A scan of common directory aggregators and submission services reveals listings that may have been submitted by predecessors or third parties. A query to companies-house-style registries reveals official listings. A cross-reference against AI citation tracking reveals listings that AI systems are aware of, including some the company itself was unaware of.
The inventory should record, for each listing: directory name and URL; date of last update; current description used; categorisation; contact details listed; any premium features active; cost of inclusion (annual or one-time); and the contact path for the directory’s operator. This data forms the basis of subsequent SIGNAL scoring and, equally important, of the maintenance plan that follows.
An audit conducted with rigour reveals the unmanaged consequences of years of unsystematic submissions. A typical mid-sized company finds dozens of listings whose existence was unknown to current staff; multiple distinct descriptions in circulation, none of which match the current canonical positioning; outdated office addresses on perhaps 15-25% of listings; categorisations that no longer reflect the company’s product line; and several paid listings whose costs are being incurred without anyone tracking them. Each of these findings is both a problem to remedy and a chance to consolidate around a coherent directory strategy.
Findings from research published by the OECD on digital data infrastructure (~2023) emphasise the importance of consistent metadata across data sources for accurate downstream use; the same principle applies to directory listings. Inconsistency across listings is not merely cosmetic; it actively degrades the ability of AI systems to resolve the company as a coherent entity, which in turn degrades citation accuracy. The audit must therefore produce both a list of directories and a list of inconsistencies to remediate.
Prioritizing new submissions
After the audit, the team typically identifies a list of directories where the company is not currently listed but should be, drawn from competitor analysis, citation tracking, vertical knowledge, and gap analysis against the SIGNAL portfolio target. Prioritising these new submissions means balancing several factors: the directory’s SIGNAL score, the difficulty of inclusion (free submission, paid inclusion, editorial review, institutional affiliation), the time-to-visibility-impact (some directories surface listings within days, others within months), and the maintenance burden once included.
A reasonable prioritisation logic ranks new submissions by expected visibility lift per unit of effort. High-SIGNAL directories with low-effort inclusion go first; high-SIGNAL directories with high-effort inclusion go second (because the effort pays off, but takes longer); medium-SIGNAL directories with low-effort inclusion go third (quick wins that cumulatively matter); medium-SIGNAL directories with high-effort inclusion go last or are deferred. Low-SIGNAL directories are excluded regardless of effort, per the threshold logic.
The prioritisation should also consider sequencing dependencies. Some high-authority directories require evidence of inclusion in other directories or in trade press as a precondition for review; the lower-tier inclusions may need to come first to enable the higher-tier ones. The team should map these dependencies explicitly, rather than discovering them on submission. A typical sequence might involve securing inclusion in two or three trade-association directories first, turning that institutional recognition into press coverage, and then using both as supporting evidence for inclusion in a more selective analyst-aligned directory.
Submission quality matters more than submission speed. A rushed submission to a high-quality directory often results in rejection or in a poor-quality listing that the company then has to maintain. Each high-priority submission should be treated as a small project: canonical description prepared, supporting documentation gathered, contact path clarified, follow-up planned. The Deloitte Insights observation that most skills-based talent strategies do not deliver value because they begin with skills rather than outcomes applies here too: directory submissions that begin with the form fields rather than with the desired outcome tend to produce listings that match the form but not the goal.
Measuring AI visibility lift
The final implementation discipline is measuring AI visibility lift in a way that informs ongoing portfolio decisions. Measurement should be designed before the portfolio changes are made, so that baseline data exists to compare against. The temptation to skip baseline measurement is significant, since teams want to begin work immediately, but the cost of skipping is permanent ambiguity about what the work produced.
The baseline should capture: AI citation frequency for a defined basket of representative queries across major AI systems (Perplexity, ChatGPT, Claude, Gemini, Google AI Overviews); citation accuracy metrics for those citations; competitive share-of-voice in the same query basket; and any traffic or conversion data attributable to AI referrals. The query basket should be stable over time; changing the basket between baseline and follow-up measurements destroys comparability. Baseline data should be captured at multiple points before changes begin, to establish natural variability.
Follow-up measurement should occur on a regular cadence, monthly or quarterly, and should distinguish between query-level changes (specific queries where citations changed) and aggregate changes (overall trends across the basket). Aggregate trends are more reliable; query-level changes are more diagnostic. A query that previously cited a competitor but now cites the company is interesting; a basket-level shift in share-of-voice is decisive.
The measurement framework should also separate causation from correlation. AI systems change their retrieval behaviour for reasons unrelated to any given company’s directory portfolio; broader updates to the underlying models, changes in retrieval algorithms, shifts in source weighting, and competitor activity all produce signal changes that have nothing to do with the focal company’s actions. Ascribing all observed changes to portfolio actions overstates the portfolio’s effect. A reasonable approach is to measure relative share-of-voice rather than absolute citation frequency, since relative measures partly control for ecosystem-wide changes.
Findings from Wiley’s editorial guidance on search visibility indicate that over half of Wiley Online Library traffic comes directly from search engines like Google and Google Scholar; the analogous metric for AI visibility, what proportion of relevant inbound interest is mediated by AI citations, is a useful long-term indicator for any directory programme. The Wiley figure is a reasonable benchmark for what is achievable when content is well-structured for machine consumption; companies whose AI visibility is well below this proportion likely have either content quality problems or directory portfolio problems, and SIGNAL helps diagnose which.
Measurement findings should feed back into the framework iteratively. After two or three measurement cycles, the team should review which directories are over-performing relative to their SIGNAL scores and which are under-performing, and adjust scoring or weighting accordingly. The framework is a starting point for disciplined decision-making, not a static rulebook; the empirical observations from sustained measurement are what eventually produce a portfolio tuned to the specific company’s situation. World Bank research on industrial policy notes that comprehensive policy frameworks involve fifteen distinct policy tools beyond the traditional focus on tariffs and subsidies; the analogous insight for visibility frameworks is that the dimensions captured in SIGNAL are themselves a starting set, and additional dimensions may need to be added for specific verticals or specific company circumstances. A framework that cannot evolve cannot remain useful.
One caveat about measurement deserves emphasis. AI citation behaviour is currently volatile because the underlying systems are evolving rapidly. A measurement programme that produces stable, comparable data across a year is rare; most programmes have to absorb several methodological adjustments as the AI landscape shifts. Teams should plan for this volatility rather than be surprised by it; the goal is not perfect comparability but useful directional signal. The Deloitte observation about synthetic content flooding the workplace, and the resulting question of how leaders discern what is real, applies to measurement too: the measurement infrastructure must itself be robust to the evolving content landscape, which means using multiple independent measurement sources and triangulating across them rather than relying on any single tool.
Looking ahead over a three-to-five-year horizon, several trends seem likely to shape the directory landscape and the SIGNAL framework’s application to it. First, retrieval-augmented generation will become more sophisticated about source weighting, which will favour high-authority directories with editorial standards and disadvantage low-quality aggregators; this will widen the quality gap that SIGNAL already reflects, and the cost of including in low-quality directories will rise as those directories are increasingly filtered. Second, knowledge graph linkage will become more important as AI systems standardise on canonical entity identifiers; directories that participate in linked-data initiatives will gain visibility share at the expense of those that do not. Third, vertical directories operated by trade bodies and academic institutions will gain authority weight relative to commercial directories, because their non-commercial governance is more legible to retrieval systems applying credibility filters. Fourth, the emergence of category-specific AI search interfaces, vertical AI search engines for legal, medical, technical and financial queries, will create new directory contexts whose selection and weighting differ from general-purpose AI systems; SIGNAL will need a vertical-AI-specific extension to handle them.
The prediction that flows from these trends: by 2028, the directories that produce the largest visibility lift will be a small minority of the directory landscape, perhaps the top 10-15% by quality, and the cost of being outside this top tier will be effectively complete invisibility in AI-generated answers. The prediction holds if retrieval systems continue their current trajectory toward source-quality filtering and if vertical AI search becomes a meaningful share of total query volume. The prediction would be falsified if AI systems regress toward broader retrieval (treating more sources as eligible citations) or if a regulatory framework emerges that mandates source neutrality in AI citations, neither of which appears likely on current evidence, but neither of which is implausible. Practitioners building directory portfolios today should plan for the more concentrated future while keeping enough optionality to adapt if the trajectory shifts.
