The AI local search story is really a review data story

For the last eighteen months, the local SEO conversation has been dominated by AI Overviews, ChatGPT citations, and the general anxiety about what happens when nobody clicks through to a Google Business Profile any more. GatherUp's new research, presented by Annie Jackson and Jason Wertham this week, quietly reframes the whole thing.

A 3.3-star car wash won an AI answer over higher-rated competitors because it matched a specific query about SUV clearance and 24/7 hours. That's the headline example. But the more consequential finding is buried underneath: the major directory platforms — Google, Yelp, and the rest — block LLM crawlers from reading review content on business listings. Which means the reviews you've spent years accumulating aren't feeding AI answers about your business at all. Not directly. Not from the source you assume they are.

That's a bigger deal than the industry has processed.

The directories are a walled garden the LLMs can't enter

Wertham's framing is worth reading twice. When ChatGPT or Claude answers a question about a local business, it isn't scraping the Google reviews. It can't. The directory owners have blocked LLM crawlers from the review content that sits inside their profiles. What the LLM has access to is whatever review content lives on the open, crawlable web — republished testimonials on your own site, embedded review widgets, screenshots on social, quotes in press coverage, mentions on Reddit and forums.

The implication is that two businesses can have wildly different Google review profiles and produce identical AI answers. Because the AI isn't looking at the Google reviews. It's looking at the shadow of your reviews — the fraction that leaked out onto the open web.

Most local businesses have never thought about their reviews this way. They've been optimising for a Google surface that's becoming less relevant to the answers customers actually see, whilst the surface that does feed those answers — their own website, their social channels, third-party mentions — is thin, unmanaged, or empty.

Query specificity beats star rating, and that changes the optimisation target

The car wash example is the part that will make local SEOs uncomfortable. A 3.3-star business won because the AI answer prioritised query match over aggregate rating. The customer asked about SUV clearance and 24/7 hours. The AI found a business whose listings and reviews contained those specifics. The star rating was secondary.

The AI isn't ranking businesses. It's matching descriptions to questions.

The AI isn't ranking businesses. It's matching descriptions to questions.

This is a subtle but important shift. Traditional local SEO logic says: get more reviews, get better reviews, move your average rating up, watch your rankings improve. AI-answer logic says: get reviews that mention the specific things customers ask about, in language customers actually use, in places the LLM can read.

A dozen five-star reviews saying "great service, would recommend" are worth almost nothing in this new context. Two three-star reviews mentioning "took my Range Rover through with room to spare" and "open at 2am when I needed it" might win the answer.

Which means the reviews that matter aren't the ones that lift your average. They're the ones that populate the LLM's understanding of what your business actually does, for whom, under what conditions.

The first-party review capture problem just got worse

There's a long-standing debate in local marketing about first-party review capture — the surveys and follow-up emails that generate honest customer feedback but never make it onto public platforms. The traditional argument against them was straightforward: reviews that only you can see don't help you rank. So businesses either pushed customers to Google or gave up on structured feedback entirely.

The AI-search context makes first-party capture even more useless than before. Reviews that sit in your CRM don't help you rank in Google, and they don't feed AI answers either. They're private data with no distribution.

But the same context makes republishing that first-party feedback dramatically more valuable. If you own the content, you can put it on your website, in your social feeds, in your press outreach, in your podcast transcripts. Every one of those surfaces is crawlable. Every one of them contributes to the LLM's picture of your business. The reviews you weren't using publicly are the raw material for the AI-search layer you don't currently have.

Most local businesses have thousands of these sitting unused. Wertham mentioned one customer with 11,000 first-party responses that had never been surfaced anywhere the LLMs could read. That's a distribution asset the business already owns and has never deployed.

Context signals I didn't realise were baked in yet

The other detail that deserves more attention is Wertham's comment about temporal and personal context. Time of day is now a ranking signal in Google Maps. If the LLM knows you own an SUV, or a large dog, or that you tend to search at 11pm, those signals apply to every subsequent local query — restated or not.

This has been theoretically possible for years. What's changed is that it's now happening in production, on queries local businesses can't see or intercept. You can't optimise for a query the customer never fully articulates because the AI has already inferred the missing context.

The response isn't to try and reverse-engineer the personalisation. It's to make sure your business description — across every crawlable surface — is specific enough that the AI can match it to a wide range of inferred contexts. Vague descriptions lose. Detailed, granular, use-case-heavy descriptions win.

That's the loop. And most local businesses have built the opposite of it — sanitised, keyword-driven copy that says nothing specific about who they serve or when.

What this actually means if you run local pages

If you're a multi-location business, or you run local SEO for one, the immediate action list looks nothing like the traditional local SEO checklist:

The reviews on your Google profile are load-bearing for the Google surface but structurally invisible to LLMs. You have to republish the useful ones somewhere crawlable — your own site, ideally with structured markup, but also social channels and any earned press. The 11,000-response drawer of first-party feedback is a distribution asset, not a private data set. Categorise it, extract the specific-use-case testimonials, publish them.

The reviews you want more of aren't the ones that lift your star rating. They're the ones that answer the specific questions customers ask AI. If the AI is being asked about SUV clearance, dog-friendly patios, late-night availability, wheelchair access — the reviews that mention those things are the ones that win answers. That means prompting customers post-purchase in ways that surface specifics, not generalities.

And the location pages themselves need to stop reading like generic templates. If every location page on your site says the same thing with the city name swapped, you've built content that's invisible to the AI matching layer. Location pages have to carry the specific context that would let an LLM answer a specific question about that specific site.

The honest limit

None of this replaces traditional local SEO. Google still returns Maps results. Star ratings still influence conversion. Google reviews still matter for the Google surface, which is still the majority of local discovery for most businesses. What's changed is that a growing minority of local queries — GatherUp's data says 55% of consumers have used AI summaries, 48% have asked ChatGPT — now runs on a completely different substrate. And that substrate reads a completely different set of signals.

The businesses that will win the AI-answer layer aren't the ones with the best Google profiles. They're the ones whose reviews, descriptions, and context signals live in places the LLMs can actually see. Right now, that's almost nobody.