Your Buyer Isn't on the Engine
You're Ranking On.

A marketing director at a heritage hotel group runs a quarterly report. The agency dashboard is reassuring. #1 for “luxury Cotswolds country house hotel.” #2 for “boutique hotel Oxfordshire.” Top three on every commercial query that matters. The board nods. The retainer is renewed.

Three weeks later, an enquiry from a guest planning an anniversary mentions, in passing, that they had asked ChatGPT for a shortlist. The hotel was not on it. The marketing director makes a note to ask the agency. The agency replies that ChatGPT isn't in their tracking remit; it's “an emerging channel.” The conversation ends.

What the marketing director has just brushed against, without naming it, is the central change in commercial visibility since PageRank. The shortlist their buyer is being given is not generated from the index they have been ranking on. It is generated from a different index entirely. The rankings on the agency dashboard are real. They are also, increasingly, irrelevant to the conversation that decides whether the buyer ever hears the brand's name.

This is the Multi-Index Problem. And almost no UK marketing leader has been shown the shape of it yet.

The architecture of AI search is not what most marketing directors assume. The assistants buyers are now using do not all read the same web. They read different webs, curated by different crawlers, ranked by different rules.

The breakdown matters because the implications cascade from it.

Google's family (Gemini, Google AI Overviews, Gemini in Workspace, Apple Intelligence's web fallback) reads the Google index. This is the index every marketing director has been optimising for since 2008. It is comprehensive, sophisticated, and the home of the agency dashboard. It is also, increasingly, only one of three.

Microsoft and OpenAI's family (ChatGPT, Microsoft Copilot, DuckDuckGo's chat) reads a Bing-foundational index. Bing supplies the underlying retrieval; OpenAI's own search crawler layers proprietary indexing on top. ChatGPT, by usage, is now the most-asked AI assistant in commercial contexts. The marketing director who has not claimed Bing Webmaster Tools is, functionally, hoping that what Bing happens to know about their business is accurate. As reported elsewhere in this series, research happens in AI; shortlisting happens in AI; Google is increasingly the validation step, not the starting point. For the family of assistants reading Bing, that mid-funnel reality is felt before any Google ranking is consulted.

Anthropic's family, Claude, is widely evidenced to rely on Brave Search for web retrieval. Brave is the smallest of the three by raw scale, but is the only major web index built independently of Google and Microsoft. It indexes differently. It ranks differently. It values different signals. A brand that is well-represented on Google can be functionally invisible on Brave. A marketing director who has never run their brand name on search.brave.com cannot tell you what Claude believes about their business. Most cannot.

Two further engines round out the picture. Perplexity runs hybrid retrieval (its own crawler-built index plus third-party search providers), with a strong bias toward sources it considers authoritative: academic publications, established trade press, named experts. Brave's own assistant Leo reads the Brave index natively. Smaller volume, more strategically interesting in privacy-conscious B2B contexts.

The Multi-Index Problem, as set out above, is the gap between Google, Bing, and Brave. That gap is the floor of the problem, not the ceiling. Even the same company, querying its own index for the same buyer asking the same question, returns different answers depending on which surface the buyer chose to open.

To test it cleanly, we ran a single query; “I’m getting married and I want a big party; top 5 wedding venues in the Cotswolds”; through three Google products, within minutes of each other.

★   named in all three Google surfaces

Three Google products. Three different shortlists. Only one venue (Manor By The Lake) appears in all three. Gemini, the assistant, shares only that one venue with the other two surfaces. Search’s AI Overview and AI Mode share three (Stone Barn, Manor By The Lake, Blenheim) but disagree on the remaining two. Two new venues enter the conversation that none of our four cross-engine tests named at all: The Frogmill, surfaced only by AI Overview, and Cotswolds Hotel & Spa, surfaced only by AI Mode.

The implication for the marketing director’s mental model is significant. The standing assumption; “if we’re indexed by Google, we’re visible across Google’s products”; is no longer true. Each surface applies its own retrieval bias, ranking weights, and sourcing rules on top of the shared index. The shared index is not the shared answer; it is the input. The AI Overview prioritises differently from AI Mode. AI Mode prioritises differently from Gemini. Gemini, in turn, has its own opinionated retrieval logic that produces a list almost no other Google surface confirms.

This loops directly back to the central argument of section II: the marketing director who treats “Gemini visibility” as a stand-in for “Google AI visibility” is making the same category error a buyer makes when they treat one assistant’s answer as the buyer’s truth. There are at least three Google surfaces, with at least three different definitions of who matters. Tracking only one of them is, again, watching the smallest screen in the room.

The Multi-Index Problem is not solved by panic, and not solved by hiring three agencies. It is solved by changing what the quarterly report measures, and changing the standing assumption that “we rank well on Google” is the answer to the visibility question.

Three shifts, none of them expensive, sit at the heart of the discipline.

The first is to broaden what “visibility” means in the standing report. A quarterly marketing report in 2026 that contains rankings only from Google is reporting on one third of the buyer's reality. The rest of it (what ChatGPT says when asked the brand's category question, what Claude says when asked for a recommendation, what Perplexity cites when asked for a comparison) is not in the report not because it cannot be measured, but because nobody has asked for it to be. Those answers are checkable in five minutes, by anybody, on any laptop. Most marketing directors have never run them.

The second is to claim the ground that has been silently abandoned. Bing Webmaster Tools is free and takes ten minutes to set up. Most UK businesses we audit have never claimed it, which means ChatGPT and Copilot are reading whatever Bing has scraped automatically, with no input from the brand. search.brave.com is publicly searchable in three seconds. Most marketing directors have never tested it. These are not budget questions. They are noticing questions.

The third is to retire “we rank #1 on Google” as a statement of marketing health. The line was adequate for a decade. It is no longer. Not because the Google ranking has stopped mattering (it has not), but because it is now one signal among at least three, and the other two are growing faster than the first.

The old line was a status statement, checked once per quarter and written into a board pack. The new line is an ongoing question, checked every time a model retrains. A status statement says we are visible. A diagnostic asks are we still visible, and on which surfaces? The first you report. The second you operate.

A marketing director presenting Google rankings to the board in 2026 without context for ChatGPT, Claude, and Perplexity visibility is presenting an incomplete picture. The board will not always notice the gap. The buyer is already in it.