What this covers

• The Inputs Are Not the Same
• The Bing Dependency
• Mentions Without Links
• Reviews Are Read, Not Counted
• Forums Outrank Company Websites
• Structured Data Is Doing More Work Than It Was
• The Measurement Problem Nobody Has Solved
• What to Do First
• What the Standard Advice Misses
• The Summary

A business can occupy the top of Google for its main commercial term and be entirely absent when somebody asks an AI assistant the same question in plain language.

This is not a bug in either system. They are different systems, drawing on different inputs, and a decade of optimization aimed at one of them produces very little advantage in the other.

For anyone whose customers have started asking questions conversationally rather than typing keywords, that gap is now a commercial problem rather than a curiosity.

The Inputs Are Not the Same

Large language model assistants draw on search index data, on what was in their training data, and increasingly on live retrieval performed at the moment of the question.

Backlinks carry less weight in AI retrieval than in classic search. This is the single most consequential difference. Link authority is the load-bearing signal in Google’s organic ranking and the thing most SEO budgets are ultimately spent on. An assistant assembling an answer about local businesses is not running a link graph calculation.

What it is doing looks much more like reading. It draws on text about the business, wherever that text lives, and assembles an answer from it.

The right column is close to being the inverse of a conventional SEO checklist, which is why businesses that have invested heavily in one column are frequently absent from the other.

The Bing Dependency

ChatGPT retrieves live results through Bing. That single technical fact has an immediate practical consequence that most local businesses have never acted on.

A business with no Bing Places listing is missing from the index that the most widely used assistant queries when it needs current local information. Not ranked poorly. Missing.

Creating that listing is free, takes under an hour, and is the highest-return single action available in this area. It is also skipped almost universally, because Bing’s own search share made it look irrelevant, and that reasoning was sound right up until it was not.

The same logic extends to the other major directories. Yelp, the Better Business Bureau and Apple Maps all carry business information into places assistants read, and they matter for reasons that have nothing to do with the traffic they send directly.

Mentions Without Links

An unlinked brand mention carries weight in AI retrieval. This is a genuine reversal of a long-standing assumption.

Classic link building treats an unlinked mention as a missed opportunity, and outreach campaigns exist specifically to convert mentions into links. In AI retrieval the mention itself is the asset. The business name appearing in relevant text, in a credible place, is what makes it available to be recalled and cited.

That changes what a public relations placement is worth. A piece in a local publication that names a business without linking to it was previously a modest brand outcome and no SEO outcome. It is now a direct input to how an assistant answers questions about that category in that city.

It also changes the value of being written about at all, which is a different activity from being linked to and requires different work.

Reviews Are Read, Not Counted

Review text provides matchable detail. An assistant answering a specific question is looking for text that addresses it.

A business with two hundred reviews saying “great service, highly recommend” has an excellent rating and almost nothing an assistant can use to answer a specific question. A business with forty reviews that describe particular circumstances, particular services, particular problems solved, has a body of text that matches specific queries.

The practical implication for review generation is uncomfortable, because the standard approach optimizes for the wrong thing. Asking for a rating produces ratings. Asking a customer what the job was and how it went produces text, and the text is what gets read.

Platform diversity matters here too. Assistants draw from multiple review sources rather than exclusively from Google, which makes a business with reviews in one place narrower than its rating suggests.

Forums Outrank Company Websites

Community forums are frequently cited by AI assistants, often above the websites of the businesses being discussed.

The reason is that assistants weight what appears to be independent testimony above what is obviously self-description. A company page saying it is reliable is a claim. A thread in which several people discuss their experience is evidence, and it reads as such.

There is an obvious temptation attached to that observation and it is worth naming as a mistake rather than a tactic. Communities detect and punish marketing, and an account posting promotionally is removed along with everything it contributed. The only version that works is genuine participation over time by someone who actually knows the subject, which is slow and is not a campaign.

For a business with real expertise, that is a lower barrier than it sounds. The person who does the work already has the answers. What is missing is the habit of answering in public.

Who does the work is the substance behind all of it, which is why the team behind our local search work is a page rather than a formality, and the agency profile carries the same information where an assistant is more likely to read it.

Structured Data Is Doing More Work Than It Was

Schema markup describes a business in structured form. It has always been useful and it has become considerably more so, because a system assembling a factual answer benefits from facts already labeled as facts.

The practical version is unglamorous: organization markup site-wide, local business markup on the location page, service markup on service pages, FAQ markup where questions are genuinely answered, and review markup where reviews are displayed. Validated, rather than assumed to be working.

Most sites either lack it or carry a partial implementation generated years ago by a plugin and never checked since. Both are common and both are cheap to fix.

The Measurement Problem Nobody Has Solved

There is no equivalent of a rank tracker for this.

Search Console does not report AI Overview appearances. No tool reliably reports whether an assistant recommended a business, because the answer varies by phrasing, by session, and by user context, and there is no impression log to read.

The available approaches are all partial. Asking assistants a set of consistent questions periodically and recording what comes back is manual and directionally useful. Brand monitoring tools have started reporting AI visibility with real limitations. Neither is measurement in the sense that rank tracking was.

The correct response is to say so plainly rather than to invent a metric. Work in this area is currently justified by reasoning about the mechanism rather than by a dashboard, and clients deserve to be told which of those they are being sold.

What to Do First

Ordering the available work by return rather than by novelty.

The first three are one-off tasks that most businesses have never done, cost nothing beyond time, and are complete within a week.

The bottom row is included because it is where local budgets currently go. Links remain valuable in classic search and they are close to irrelevant to the question of whether an assistant recommends a business, which makes a link-only program a bet on one of the two systems.

What the Standard Advice Misses

Most published guidance on this subject treats AI optimization as an extension of SEO, with a few additions bolted on. The signal weights are close to inverted, which makes it a parallel discipline rather than an extension.

The second omission is the Bing dependency. It is the most actionable fact available, it costs nothing to act on, and it is buried in most coverage beneath advice about content quality.

The third is the measurement gap. A great deal of writing on this topic implies results can be tracked in the usual way. They cannot yet, and pretending otherwise sets up exactly the disappointment that follows any promise nobody can verify.

The Summary

AI assistants and search engines read different signals. Links, exact-match domains and page speed carry little weight in AI retrieval. Mentions, review text, structured data and community discussion carry a great deal.

ChatGPT reaches live local information through Bing, which makes a free Bing Places listing the highest-return action available and one almost nobody has taken.

Reviews are read rather than counted, so specific text beats a high count. Forums are cited above company websites because independent testimony reads as evidence.

And none of it is measurable the way rank was, which is worth admitting rather than papering over.