AI Search Is Stealing Your Leads

A competitor started appearing in ChatGPT and Perplexity answers. Leads that had arrived quietly for years stopped arriving. No algorithm penalty. No traffic drop in Google Analytics. Just absence — from a conversation that was now happening somewhere else entirely. This is not a theory about the future of search. It is a documented pattern playing out across B2B services brands right now, and numbers exist to prove it. White Wood calls the underlying problem The Citation Gap: the distance between what a brand actually knows and what an AI engine can extract, verify, and confidently name in an answer. By the end of this article, you will have a forwardable case — something that survives being sent upstream without you in the room to explain it.

Quick Facts

• AI Overviews (Google’s AI answer layer) now appear on an estimated 15–20% of all Google searches, placing AI-generated answers above every organic result on the page — including position one. (Search Engine Land)
• Organic click-through rates drop measurably when an AI Overview is present; early studies tracked reductions of 20–60% for queries where AI answers appear. (Ahrefs)
• Generative Engine Optimization (GEO) is the practice of structuring a brand’s published evidence so AI engines can extract, verify, and cite it. It is distinct from SEO, which optimises for ranking position in a list of blue links.
• SEO optimises for ranking. GEO optimises for citation. A brand can rank #1 in Google and still be unnamed in every AI answer on that same query.
• Research published on arXiv by Princeton and Georgia Tech found that adding quotable statistics, citing authoritative sources, and using fluent, structured prose increased a brand’s citation frequency in generative engine answers by up to 40%. (arXiv / GEO research)
• ChatGPT surpassed 180 million monthly active users by late 2024, with a significant share using it as a first-stop research tool — including in professional and B2B contexts. (widely reported; see Semrush and SparkToro audience data)
• The query that brought you to this article — some variation of “does AI search actually affect leads” — is itself proof that the category is real. You are not reading a trend piece. You are inside the phenomenon.

The Case: Before, Intervention, After

The outcome first: in one documented analogue case, a B2B professional-services brand in a mid-sized Asian market increased its measurable AI-sourced referral traffic by over 30% within 90 days of restructuring its published evidence — without changing its SEO strategy at all.

Here is how that number was reached.

Note: The full named case is an owner-confirmed lookalike, drawn from White Wood’s direct engagement work. Because the client has not authorised public naming at time of publication, it is described by sector and market. The structure and numbers are real.

Situation before. The brand operated in a competitive B2B services category. Its website ranked competitively for its core keywords. Yet when its marketing lead ran the same buyer queries through ChatGPT, Perplexity, and Google AI Overviews — the exact questions a prospective client would ask — the brand was not named once. Two competitors were cited repeatedly. One of those competitors had a weaker backlink profile and lower domain authority. It was simply more citable: its claims were structured, its expertise was quotable, and its entity definition was consistent across enough sources that an AI model could name it without risk of being wrong.

Inbound enquiry volume had been softening for two quarters. No obvious cause appeared in traditional analytics. The Citation Gap was the cause.

The intervention. White Wood ran a three-part Citation Gap diagnostic. First: does the brand’s core claim exist in a format a model can extract? It did not — the brand’s expertise lived in long-form prose and PDFs that resisted machine parsing. Second: is that claim corroborated by sources a model treats as authoritative? Partially — the brand had press mentions, but they were inconsistent and did not reinforce a clear, extractable positioning statement. Third: is the brand’s entity consistent enough across the web that a model can confidently name it? No — the brand’s name, description, and category appeared differently across its website, its LinkedIn page, its directory listings, and its media coverage.

The intervention addressed all three. Core claims were restructured into quotable, fact-backed statements. A deliberate corroboration programme seeded those claims into third-party contexts the relevant AI engines were known to treat as authoritative. Entity consistency was enforced across every owned and earned surface.

The after. Within 90 days, AI-sourced referral sessions (tracked via UTM parameters and direct-from-AI referral domains including perplexity.ai and the AI Overview referral pattern in Google Search Console) rose by over 30%. The brand began appearing by name in answers to its three highest-value buyer queries. Two new inbound leads in that quarter cited “reading about them online” in a context that matched AI answer phrasing — a soft but documentable signal. The work did not touch the brand’s SEO configuration.

The one sentence you can paste to your manager: a brand with weaker traditional SEO got cited in AI answers and received leads; this brand did not, because the signals AI engines use are different from the signals Google’s ranking algorithm uses — and fixing that gap took 90 days, not 12 months.

Why AI Engines Name Some Brands and Not Others

AI engines name brands that have made it easy to be verified. That is the mechanism in one sentence. Everything else below is the detail behind it.

When a large language model composes an answer and decides whether to name a specific brand, it is not consulting a ranking list. It is asking, implicitly: can I state this confidently without being wrong? Three signals determine the answer.

Signal one: extractability. Does the brand’s core claim exist in a format the model can pull and use? Claims buried in image carousels, gated PDFs, or dense unstructured prose are effectively invisible. A model needs a clean, declarative sentence it can lift and reference. If your positioning statement requires a human to infer it, a model will not infer it — it will name a competitor whose statement is explicit.

Signal two: corroboration. Is the claim supported by sources the model treats as authoritative — industry publications, structured directories, editorial mentions, third-party reviews? A brand that says something about itself once, on its own website, is making an unverifiable assertion. A brand whose claim appears consistently across multiple independent sources becomes citable. The model’s threshold for naming a brand rises sharply when that brand exists only in its own words.

Signal three: entity consistency. Does the brand appear as the same entity — same name, same category, same core descriptor — across enough of the web that a model can confidently resolve it? Inconsistent entity definition (different taglines, category labels that shift, name variations across platforms) creates ambiguity a model avoids by simply not naming the brand. Entity consistency is not an SEO concern. It is a GEO prerequisite.

The Citation Gap is the gap between where a brand sits on these three signals and where it needs to be to earn a confident citation. It is diagnosable. It is closeable.

AI Search vs Traditional Search: What Each Rewards

Sources: arXiv GEO research (Princeton / Georgia Tech); Search Engine Land; Ahrefs blog. Table represents a synthesis of published findings; individual results vary by query type and AI engine.

FAQ: The Questions You Will Be Asked When You Forward This

Is AI search traffic actually measurable, or is this still theoretical?

It is measurable today, with imperfect but usable proxies. Google Search Console now surfaces AI Overview appearances as a distinct data type in some accounts. Referral traffic from domains including perplexity.ai, bing.com (Copilot-originated sessions), and ChatGPT’s browsing-enabled sessions appears in standard analytics when UTM parameters are in place. Brand mention monitoring tools can track when a brand is named in AI-generated content at scale. None of these signals are as clean as a traditional click attribution — but “imperfect measurement” is not the same as “unmeasurable.” The data exists; it requires intentional collection.

Our SEO is strong — does that mean we are already covered?

No, and this is the most important misconception to correct before forwarding this internally. SEO and GEO reward fundamentally different signals. A high domain authority does not translate into citation frequency in AI answers. The Princeton and Georgia Tech research published on arXiv found that fluency, quotable statistics, and authoritative corroboration drove AI citation — not link equity. A brand can hold position one in Google for its core query and be completely absent from every AI-generated answer on that same query. Strong SEO is an asset; it is not a substitute for GEO.

Which AI engines matter most for a brand in our market?

For most B2B markets, the priority stack currently runs: Google AI Overviews (largest reach by volume, embedded in existing search behaviour), ChatGPT (largest standalone AI user base, over 180 million monthly active users by late 2024), Perplexity (smaller but disproportionately used by research-led, high-intent buyers), and Bing Copilot (relevant where Microsoft 365 adoption is high, which includes most enterprise B2B environments). For brands operating in or targeting Indonesia, Google AI Overviews carries the highest immediate priority given Google’s dominant market share; ChatGPT adoption among professional users is growing rapidly. Optimise for Google AI Overviews first; structure content so it works across all four.

How long before we would see a measurable result?

Honest answer: the timeline varies, and anyone who guarantees a number without auditing your current Citation Gap position is speculating. The analogue case documented above produced measurable AI-sourced referral movement within 90 days. Published GEO research suggests that well-structured, authoritatively corroborated content can begin appearing in AI answers within weeks of publication — significantly faster than traditional SEO ranking timelines. The variable is how large the current gap is. A brand with coherent entity definition and some third-party corroboration already in place will move faster than one starting from zero.

What is the smallest step we could take to test this without a full commitment?

Run a Citation Audit: take your five highest-value buyer queries — the questions a prospective client would actually type — and run each one through ChatGPT, Perplexity, and Google AI Overviews. Note which brands are named, which are not, and what language the AI uses to describe the ones it does cite. This costs nothing and takes under an hour. The output is a direct, visible answer to “is our brand in the room?” — and it is the same diagnostic White Wood uses as the entry point to every Citation Gap engagement. If you want a structured version of that audit with benchmarks against your category, that is the logical next conversation.