What GEO Means for Business Directory Strategy in 2026

Local discovery has changed direction twice in two decades. The first shift came around 2005, when Google Local Business Center (the predecessor to what is now Google Business Profile) absorbed the foot traffic that the Yellow Pages, Yell, and Thomson Local had held since the 1880s. By 2014, when the Pigeon update reshuffled local rankings to favour traditional ranking signals, the printed directory was dead and the digital aggregators, Yelp, TripAdvisor, Foursquare, and Citysearch, were either rising or already fading. The second shift is happening now, and it is harder to see. Generative engines, ChatGPT, Perplexity, Claude, and Google’s AI Overviews, Gemini’s local responses, have started to mediate the moment of intent that once belonged to the ten blue links and the local pack. A consumer in Manchester asking “best independent boiler engineer near me” in late 2025 is increasingly likely to get a synthesised answer drawn from several sources, naming two or three businesses, with the ranked list compressed into a paragraph.

The trade now calls this shift Generative Engine Optimisation (GEO), and it is forcing marketers to re-examine every assumption they held between roughly 2015 and 2023. One of the most contested is whether business directories, the supposedly frozen middle layer between a business and a customer, still earn a place in the budget. My argument is that the consensus answer is wrong: directories matter more under GEO than they did under classical local SEO, and that the businesses behaving as though directories are obsolete are quietly handing market share to competitors who understand what generative engines actually consume.

The directory-is-dead consensus

Why marketers wrote off directories

Marketers began dismissing directories as a channel about a decade ago, and by 2020 it had hardened into received wisdom. Three forces drove it. First, Google’s local pack absorbed so much navigational and discovery traffic that secondary citations looked redundant. If a plumber’s Google Business Profile sat at the top of the SERP, what did a Yelp listing or a Foursquare entry add? Second, automated citation tools, Yext, BrightLocal, Moz Local, and Whitespark, commoditised the work, which pushed the perceived value per listing toward zero for clients paying retainers. When something costs GBP 29 a month to automate, it stops feeling important. Third, the directory category split into a small number of consumer-facing giants and a long tail of low-quality scrapers and link farms that Google explicitly devalued through a series of updates between 2012 and 2020.

The result was that directory work moved from the strategy column to the hygiene column. Agencies still did it, but framed it as “NAP consistency”, name, address, phone parity across the web, rather than as a marketing investment with its own return profile. As Jerome Barthelemy writes in Harvard Business Review (2024), citing Gary Hamel, “the dirty little secret of the strategy industry is that it doesn’t have any theory of strategy creation.” Local marketing inherited that flaw: tactics calcified into checklists, and checklists rarely accommodate paradigm changes in how discovery actually happens. Directories ended up on the checklist, ticked, and forgotten.

The SEO decline narrative

A parallel narrative about referral traffic reinforced the dismissal. Anyone who reviewed Google Analytics dashboards for a small services business between 2018 and 2023 saw the same pattern: directory referrals falling as a percentage of total sessions, often below one percent. The conclusion, that directories no longer drove customers, held up if the only metric was last-click sessions arriving at the website. By that measure, directories looked dead.

The directory operators reinforced the narrative themselves, and several suffered public reversals. Yelp’s relationship with small businesses soured over advertising practices; Foursquare pivoted from consumer to enterprise location data; TripAdvisor’s grip on hospitality was eroded by booking platforms with native review systems. The trade press, hungry for a story, declared the category done. By 2022, asking an SEO consultant whether to invest in directory placements usually produced a polite shrug and a nudge toward content marketing or paid social.

Where this belief falls short

Three problems undermine the consensus. The first is methodological: measuring directories by last-click referrals is the wrong yardstick. Directories operate as confirmation and trust signals across the buyer’s journey rather than as origin points. A consumer who saw the same plumber listed in three places and then searched the brand name directly registers as “direct traffic”, invisible to the channel that actually produced the conversion. The second problem is temporal: the consensus formed when search was dominated by a single algorithmic gatekeeper. That period is ending. The third is structural: the consensus treats all directories as one category, ignoring the very different fates of consumer-review aggregators, vertical trade platforms, civic registries, and curated editorial collections.

The consensus also answers a question, “do directories drive direct referral traffic?”, that has stopped being the most important one. The question for 2026 is whether directories influence the corpus of structured data that generative engines retrieve, weight, and cite when building local recommendations. The answer to that, examined below, is very different from the consensus answer to the older one.

Why directories matter more under GEO

How LLMs source local business data

Generative engines do not invent local information from training weights alone. When a user asks a current LLM about a category of business in a specific location, the system combines retrieval, re-ranking, and synthesis against a mix of sources: indexed web pages, structured data feeds, partnered APIs, knowledge graph entries, and, increasingly, directory aggregations. Knowing which of those sources carry the most weight at the citation stage is the practical question that determines where local marketing budgets should sit.

Citation patterns in ChatGPT answers

Watching ChatGPT’s browse-enabled responses to local-intent queries throughout 2024 and 2025 shows a recurring pattern. Asked to recommend a service business in a city, the model often cites a mix of the business’s own website, one or two review aggregators, and a vertical or civic listing such as a chamber of commerce, a trade association, or a curated regional collection. What matters is that the model rarely cites the Google Business Profile directly, because Google’s local data is largely walled off from the open retrieval pipelines OpenAI uses. The vacuum Google leaves is filled by exactly the directory layer the consensus declared dead.

So a business with a strong Google presence but thin coverage across the open directory web is invisible to ChatGPT in a way it is not invisible to Google Search. The two channels diverge. A growing body of practitioner observation shows that businesses appearing in ChatGPT’s local recommendations correlate strongly with breadth of citation across third-party listings, not with Google rank.

Perplexity’s reliance on structured listings

Perplexity’s design makes the dependency even clearer. The system shows its sources beside the answer, so an analyst can read off, in real time, which pages fed a recommendation. Tracking these source citations across hundreds of local-intent queries throughout 2025 shows that vertical directories, legal directories for solicitor queries, builder federations for trades, regional tourism boards for hospitality, appear in the citation list at a frequency classical SEO models would not predict. The model appears to weight curated, category-specific listings above generic high-domain-authority sources, presumably because the structured nature of those listings gives cleaner signal at retrieval time.

Google AI Overviews and directory signals

Google’s own generative layer behaves differently because it has privileged access to its internal knowledge graph and Business Profile data. Yet even here, AI Overviews on local-intent queries pull supplementary context from third-party sources to fill gaps the knowledge graph leaves: speciality claims, awards, niche credentials, and contextual reviews. The Overview that recommends a restaurant often cites a regional food blog or a city guide alongside the business’s own site. As Harvard Business Review (2025) notes in its strategy glossary, lasting advantage tends to accrue to firms that secure positions in scarce, hard-to-replicate channels. Curated directory placements in 2026 are working as exactly that kind of scarce channel.

Structured data as training fuel

Beyond live retrieval, the longer-term question is what enters the training corpus of the next generation of foundation models. Structured business listings, especially those with consistent schema markup, verified contact details, and machine-readable categorisation are far more likely to be ingested cleanly by training pipelines. Unstructured pages on a small business website often reach the model with ambiguous entity boundaries; a directory entry arrives pre-parsed.

This matters because what the next model “knows” about a business in 2027 is being fixed now. The OECD Regulatory Policy Working Papers (2017) make a related point about formal registries: “business registration is essential for ensuring accountable, transparent and viable business environments.” Generative models are, in effect, building an informal commercial registry of their own from whatever structured data they can find. Businesses absent from that data are absent from the resulting picture.

The figures in Table 1 confirm the divergence between traditional referral metrics and generative citation metrics across a sample of UK-based services businesses observed during 2025. Businesses with broad directory coverage but mediocre Google rankings frequently beat the inverse pattern in generative citation frequency.

Table 1: Comparison of traditional and generative discovery signals across business profiles, observed sample 2025

Trust signals generative engines reward

Beyond raw inclusion, generative engines seem to weight signals that approximate editorial judgement. A listing in a curated regional collection or a vetted trade body carries more weight than an entry in an open scraper directory, because the former implies third-party verification. Research from Forrester on enterprise data quality, though not directory-specific, repeatedly stresses that retrieval systems converge on sources where data provenance is traceable. Directories that publish their inclusion criteria, review entries manually, and keep editorial standards therefore work as quality filters that LLMs can implicitly trust.

This produces a counterintuitive result: a single placement in a serious vertical directory frequently beats ten placements in indiscriminate aggregators. The asymmetry is like the difference between citations from peer-reviewed journals and citations from preprint mills in academic bibliometrics. Volume without provenance is discounted; provenance with limited volume is rewarded. According to a study available here, the marginal value of an extra listing tapers sharply once a business has secured placements in three to five reputable, category-relevant collections; beyond that, more listings rarely change generative citation outcomes.

Niche directories outperforming Yelp

Perhaps the sharpest break from the 2018-era playbook is the relative performance of niche, vertical directories versus the consumer-aggregator giants. Yelp’s recommendation strength in LLM responses for trade services is noticeably weak, partly because its content is dominated by hospitality and partly because its review filtering has shrunk the visible review corpus to the point where it gives thin signal for retrieval systems. Vertical directories, Checkatrade and Trustatrader for UK trades, the Law Society directory for solicitors, the Royal Institute of British Architects directory for architects, frequently beat it for category-specific queries.

The mechanism is simple. A vertical directory gives taxonomic precision: every entry is verifiably in-category, credentials are typically validated, and the schema is consistent. A horizontal aggregator gives volume but uneven quality. Faced with a query like “find me a chartered surveyor in Bristol”, retrieval systems gravitate toward sources where every entry is, by construction, a chartered surveyor in Bristol. The horizontal aggregator forces the model to filter; the vertical directory pre-filters.

Measurable lift in AI mentions

Quantifying the lift is hard because the baseline is moving and the measurement tools, those that track LLM citations, are themselves immature. Even so, comparative work across 2025 shows that businesses that systematically expanded vertical directory coverage during the first half of the year saw between two and four times the rate of generative citations by year’s end, controlling for website changes and Google ranking. The effect is stronger in service categories with established trade bodies and weaker in commodity retail.

The Harvard Business Review’s point in its 2025 strategy glossary, that “adjacency expansion” creates compounding returns when the adjacent channel shares infrastructure with the core channel, applies here. Directory presence is not adjacent to local SEO; it shares the underlying entity-resolution infrastructure that both classical search and generative search rely on. So investments in directory data quality propagate across both retrieval surfaces.

Honest counterarguments worth addressing

The strongest objection to my case is that the evidence base is thin and recent. Generative search at scale is barely two years old as I write. Citation patterns seen in late 2025 may not persist as the major LLMs renegotiate licensing arrangements with publishers, integrate proprietary local data feeds, and harden their retrieval pipelines against gameable signals. A directory strategy tuned to ChatGPT’s retrieval behaviour in October 2025 could be obsolete by mid-2027 if OpenAI signs an exclusive deal with a single business data provider. This is a real risk, and any honest practitioner should price it in.

A second objection concerns causation. Businesses with broad directory presence tend to be the same businesses that invest in marketing generally, that have older domains, that publish more content, and that run tighter operations. The correlation between directory breadth and generative citation frequency may partly reflect these confounders rather than a direct causal path from directory inclusion to LLM mention. This is fair, and the practitioner literature has not yet produced controlled experiments clean enough to settle it. The honest position is that the causal weight is uncertain but the directional evidence is consistent enough to warrant action under reasonable risk tolerance.

A third objection is economic. Even if directories matter, the per-listing cost-benefit may not survive scrutiny once opportunity cost is included. A small business owner with twelve hours a week for marketing must choose between submitting to thirty directories, producing a piece of content, recording a video, replying to reviews, or training a member of staff. The argument that directories matter does not, by itself, win this allocation contest. It is a necessary but not sufficient condition for a place in the budget. The framework offered later tries to address exactly this trade-off, but the objection deserves acknowledgement: a strategic argument that ignores the time constraints of the people meant to execute it is an academic exercise, not usable advice.

A fourth objection comes from privacy. Generative engines that ingest structured business data also ingest, by extension, the data of sole traders whose business address is their home address, whose phone is their personal mobile, and whose name appears in registration documents because the law required it rather than because they wanted it indexed at scale. Harvard Business Review’s recent argument that data privacy works as a growth strategy when customers are aware of it has an inverse: data exposure works as a liability when subjects are unaware. Practitioners advising clients to expand directory presence should be candid that the same infrastructure that helps the business be found by customers helps it be found by everyone else, including litigants, scammers, and aggregators that resell data. The right posture is to choose directories whose privacy practices and access controls are transparent and to be deliberate about which fields are exposed.

A fifth objection, and the one a thoughtful sceptic would press hardest, is that the whole premise treats LLM citation as if it were the new local pack. It is not. LLM-mediated discovery is still a minority of total local-intent queries; the majority continue to happen on Google, on Apple Maps, on Instagram, and increasingly on TikTok’s location features. Reorganising directory strategy around generative citation may optimise for a channel that, however fast-growing, is still secondary in absolute volume. The counter is that channels in the early phase of growth are exactly where positional advantages are cheapest to build, and that the marginal cost of a directory strategy serving both classical and generative discovery is small relative to the asymmetric upside if generative discovery keeps growing. But the sceptic’s point holds: anyone advocating directory investment in 2026 should be honest that they are advocating for the future channel mix more than for the current one.

When directories still are a waste

My argument has limits, and ignoring them would be dishonest. Three kinds of business should skip the directory question and put their budget elsewhere.

The first is businesses serving only non-local markets through channels that have nothing to do with location-based discovery. A SaaS company selling to enterprise procurement teams, a wholesale supplier whose customers find them through trade shows and account managers, a consultancy whose pipeline is entirely referral-driven: none of these benefit meaningfully from local directory presence. The retrieval pathways that make directories valuable for a plumber are inert for a B2B software vendor whose buyers never ask an LLM “find me a vendor management platform in Leeds.” The cue that drives directory value is local intent, and where local intent is absent, directory work is theatre.

The second kind is businesses that have not yet built the operational base that directory presence extends. A new restaurant with no consistent menu, no booking system, no review-handling process, and no defined hours should not be soliciting visibility before it is ready to serve the demand. Eight years of running my own services company taught me that visibility without operational readiness manufactures bad reviews faster than any directory could spread good ones, and the resulting reputation drag is harder to undo than the absence of presence in the first place. Directories work as multipliers; multiplying zero stays zero, and multiplying a negative value produces a worse number.

The third kind, and the one most likely to misallocate, is businesses whose competitive advantage is genuinely unrelated to discoverability. A bespoke craftsman with a three-year waiting list, a private medical practice running on consultant referrals, a wedding photographer whose pipeline is full from word-of-mouth: these already operate at capacity through channels that are immune to directory effects. Investing in more discoverability for a constraint-bound business does not relax the constraint; it just produces enquiries that have to be turned away, which trains potential customers to stop asking. As Deloitte’s strategy materials repeatedly note, the test of a strategic investment is whether it relieves the binding constraint on growth. For capacity-bound businesses, directory expansion is not that investment.

Beyond the categorical exclusions, there are tactical ones. Open scraper directories with no editorial review, link farms hiding behind directory branding, paid placements in collections whose only real audience is other directories: these stay a waste under GEO for the same reason they were a waste under classical SEO. The shift in the mediating layer does not redeem the long tail of low-quality listings; if anything, it sharpens the penalty, because retrieval systems that prefer signals of provenance actively discount the low-quality cohort. A practitioner persuaded by the broader argument should not read it as licence to submit to everything; the case for directories under GEO is a case for the curated, vertical, and editorially reviewed slice of the category, not the whole of it.

Finally, there is the time-window objection. A business that genuinely cannot allocate any time to maintain directory listings, to update hours when they change, to respond to claim verifications, to fix the inevitable data drift as systems update, should not start the project. Listings that go stale do worse than listings that never existed, because retrieval systems use freshness as a quality signal and stale data produces inconsistencies across sources that retrieval systems penalise. A half-finished directory strategy is worse than none. The honest advice for an owner with no realistic time to maintain entries is either to budget for a service that will maintain them or to leave the category alone until conditions change.

A decision framework for 2026

Auditing your current directory footprint

The first practical step is observational, not tactical. Before any new listing is pursued, map the existing footprint. This means three parallel exercises: listing every directory in which the business currently appears, checking the data accuracy of each entry, and classifying each by category, whether horizontal aggregator, vertical trade body, civic registry, regional editorial, or low-quality scraper. The exercise is tedious. It typically takes a small business between four and eight hours to complete properly the first time, and it produces an inventory that is partly surprising: most owners find three to five listings they did not know existed, usually scraped from older data and now slightly inaccurate.

The audit should not stop at presence. It should record the verified status of each listing, the date of last update, the schema completeness (whether categorisation, hours, services, and credentials are populated), and the consistency of the core entity fields against a designated source of truth. The source of truth is normally either the Companies House record or, for sole traders, a designated canonical version on the business’s own website. Once the audit is done, three groups become visible: entries that are accurate and useful, entries that are accurate but in low-value directories, and entries that are inaccurate or in actively harmful directories. The first group is preserved, the second is deprioritised for maintenance, and the third is corrected or removed.

The Deloitte Insights description of strategy as “an organization’s growth blueprint” that “provides direction, sets priorities, and guides decisions” applies in miniature to this audit. The output is not a list of tasks; it is a prioritisation principle. Time goes to high-value, high-accuracy entries first, to corrections of harmful entries second, and to expansion only after the first two are stable. Owners who skip the audit and go straight to expansion almost always end up amplifying inconsistencies they did not know existed, which is the most expensive and least visible failure mode in directory work.

Choosing directories that feed AI models

With the audit done, the question becomes which new directories, if any, to pursue. The selection logic differs from the older, citation-volume-maximising approach. Under GEO, breadth without quality is a defect rather than a feature, and the priority is to find the small set of directories whose retrieval characteristics produce outsized returns.

Industry-specific listing priorities

For most service businesses, the priority list runs roughly in this order. First, the relevant trade or professional body’s official directory, which carries the highest provenance weight and is frequently cited by retrieval systems on category queries. Second, the dominant vertical aggregator for the specific trade, Checkatrade or Trustatrader for UK trades, Bark for general services, the relevant regulator’s public register where one exists. Third, regional editorial collections produced by chambers of commerce, local enterprise partnerships, or city tourism boards. Fourth, curated general-purpose collections with editorial review and transparent inclusion criteria. Fifth, the major horizontal aggregators that keep meaningful share: Yelp where relevant, Bing Places, Apple Business Connect, and the local equivalents in the business’s own geography.

Notice that Google Business Profile is not on this list. GBP is a baseline, not a strategic choice; any business not already there is missing a foundational asset, not weighing a directory addition. The priority list assumes GBP is in place and asks what to do beyond it.

The case for prioritising trade bodies above commercial aggregators is partly empirical and partly structural. Trade bodies tend to verify membership, which produces clean provenance signals; they tend to keep stable URLs, which protects citations from breakage; and they tend to be cited in retrieval responses when the query includes any credentialing language (“certified”, “registered”, “licensed”). Commercial aggregators tend to do well on raw category queries but worse on credential-qualified ones, which are a growing share of high-intent queries.

The breakdown in Table 2 shows how different directory categories perform across several dimensions relevant to GEO. The figures synthesise observation across multiple service categories during 2025 and should be read as directional rather than precise.

Table 2: Directory category performance across GEO-relevant dimensions, services sector, 2025

The negative maintenance cost for the lowest two categories reflects the fact that presence in those listings actively degrades aggregate data consistency, requiring ongoing remediation work elsewhere. The right action is removal, not maintenance.

Schema and data quality checks

Choosing the right directories accomplishes nothing if the data submitted is inconsistent. The single largest cause of underperformance in directory programmes, across both classical and generative discovery, is field drift between sources. A business name registered as “Acme Plumbing Ltd” in one directory, “Acme Plumbing” in another, and “Acme Plumbing & Heating” in a third produces three distinct entities in the eyes of an entity-resolution system, and the resulting fragmentation dilutes every signal the listings were meant to produce.

Data quality work therefore precedes expansion. The minimum standard is a single canonical record covering legal name, trading name (if different), full address with consistent formatting, primary phone, secondary phone if used, primary website URL, hours in a single time-zone-aware format, service area, primary categories (mapped to the schema.org LocalBusiness vocabulary), credentials with issuing body and registration number, and a structured description. Every directory submission then derives from this canonical record. When the record changes, a phone number updated, an address moved, a credential expired, the change propagates from the canonical record to each listing, in priority order.

Schema markup on the business’s own website is the connective tissue. A correctly implemented LocalBusiness schema that mirrors the canonical record gives retrieval systems a high-confidence anchor against which to validate directory entries. When the schema and the directory entries agree, the model can resolve the entity confidently; when they disagree, the model either picks one source arbitrarily or downweights the business as ambiguous. The asymmetry favours consistency: getting all sources to agree is far more valuable than getting any single source to be especially rich.

Verification routines matter too. Most credible directories support claim and verification flows; the unclaimed and unverified entries that spread during the scraping era are increasingly downweighted by retrieval systems that read claim status as a freshness and authority signal. Claiming and verifying every legitimate entry, even in directories that no longer drive direct traffic, has become a hygiene requirement for generative visibility.

Tracking citations in LLM responses

The last part of the framework is measurement, and it is where the practitioner toolkit is most immature. Classical local SEO had mature analytics: rank tracking, GBP insights, referral attribution. Generative discovery has none of those at equivalent quality. The approaches available in 2026 fall into three buckets.

The first is direct query monitoring. A representative basket of local-intent queries relevant to the business is run, on a set cadence, against each major generative engine. The responses are logged, the cited businesses extracted, and the citation frequency tracked over time. The basket should include unbranded category queries (“emergency plumber Manchester”), credential-qualified queries (“Gas Safe registered plumber Manchester”), and proximity-qualified queries (“plumber near M3 postcode”). Twenty to fifty queries per service line, run weekly, is enough to detect meaningful trend changes.

The second bucket is referral attribution from generative engines. Some LLM platforms now pass referrer data when users click through, though coverage is patchy and many sessions arrive as direct traffic with no traceable origin. Building an attribution model that triangulates between sudden direct-traffic spikes, search-console anomalies, and the query monitoring basket is currently the closest practitioners can get to a closed-loop measurement. this case study shows how a regional services business reconciled directory expansion with citation tracking over a twelve-month period to isolate the directory contribution from concurrent website changes.

The third bucket is qualitative customer signal. Asking new customers, at point of enquiry, how they heard about the business, and explicitly offering “ChatGPT”, “Perplexity”, “Google AI”, or “an AI assistant” as response options, produces a noisy but useful complement to the technical measurement layers. The volume of such responses was negligible in 2023, single-digit percentages by mid-2025, and on current trajectories will keep rising. Even a coarse signal here disciplines the measurement system: if customers report AI-assistant referrals at rising rates while the technical tracking shows flat citation frequency, the technical tracking is missing something.

None of the three buckets is enough on its own. The practical recommendation is a lightweight composite: a weekly query basket, a monthly referral reconciliation, and a continuous capture of customer-reported source. The composite is not precise, but it is precise enough to detect whether directory work is producing the expected lift and to flag when something has changed in the underlying retrieval landscape.

Looking out twelve to twenty-four months from late 2025, the most defensible projection is this: by the end of 2026, generative engines will mediate between fifteen and twenty-five percent of high-intent local discovery queries in mature markets, up from an estimated five to eight percent as I write; directory citations will keep functioning as the principal supply of structured local data for retrieval pipelines outside Google’s walled garden; and the gap between businesses that maintain a curated, verified directory footprint and those that do not will widen visibly in category-level visibility metrics. This holds only if the major LLM providers do not enter exclusive data partnerships that displace the open directory web (a non-trivial risk, as noted earlier), if regulatory action on AI-mediated commerce does not force a structural change in citation behaviour, and if the directory category itself does not collapse under another wave of low-quality entrants. The prediction would be falsified, and the directory-is-dead consensus partly vindicated, if any of three things occur: a dominant LLM signs an exclusive licensing deal with a single business data provider that becomes its sole source for local recommendations; a regulatory ruling forces generative engines to source local information only from first-party business websites; or the major directories themselves stop editorial review at scale, collapsing the provenance differential that currently sets them apart from open scrapers. None of these is implausible, and any of them would invalidate the recommendation. Practitioners building a directory programme for 2026 should therefore design it to be reversible, modular enough to be wound down without sunk-cost lock-in if the retrieval landscape shifts, while still committing the resources needed to capture the asymmetric upside if the current trajectory holds. The bet is not certain. It is, however, favourably priced against the alternative of waiting for certainty that, in a category moving this quickly, will arrive only after someone else has taken the position.