How ChatGPT and Google's AI choose which local businesses to recommend

ChatGPT, Perplexity, Claude and Google’s AI features do not keep a list of the best plumber in Almada. When someone asks, they search: they pull web pages, map listings, ratings and directory entries, then write an answer from what those sources agree on. To be recommended, your business has to be findable by their crawlers, described consistently across the sources they read, and easy to quote on your own site.

That is the mechanism. The rest of this article covers what each piece means in practice and how to measure it without fooling yourself.

Step 1: the assistant retrieves, then writes

For a local question (“good physio in Barreiro”, “vet open Saturday in Seixal”), an assistant with web access typically:

1. turns the question into one or more searches;
2. fetches results from a search index, and for places, from map and review data;
3. reads a handful of those sources;
4. writes an answer naming the businesses that come up most clearly and most consistently.

The words “most consistently” matter. An assistant tends to repeat what several independent sources say about you. If your Google profile, your website, Facebook and two directories all say the same name, the same address, the same services and a similar rating, you are an easy, low-risk recommendation. If they disagree (an old phone number, a different business name, a service you stopped offering), you are a risky one.

What one real answer looked like

On 17 August 2026 we asked ChatGPT for “melhores clínicas de estética facial em Lisboa” (best facial aesthetics clinics in Lisbon). The answer listed four clinics, each shown with a Google star rating and review count, next to a map.

That is one observation, not a rule. Answers vary by wording, by day and by user. But it shows the kind of data the answer was built from: map listings and review counts, not just website copy. A thin or neglected Google profile was not going to appear in that answer, however good the website was.

Step 2: be reachable by the right crawlers

Each AI company runs separate crawlers for search/retrieval and for model training. Blocking the wrong one is the most common technical mistake we find. Here is what the official documentation says.

Company	Used for search / answers	Used for training	What blocking it does
OpenAI	OAI-SearchBot (indexes for ChatGPT search); ChatGPT-User (fetches a page when a user asks)	GPTBot	Opting out of OAI-SearchBot means the site “will not be shown in ChatGPT search answers”. Disallowing GPTBot signals content should not be used for training.
Anthropic	Claude-SearchBot (indexes for search); Claude-User (fetches for a user’s question)	ClaudeBot	Blocking Claude-SearchBot or Claude-User may reduce visibility in Claude’s answers; blocking ClaudeBot excludes future content from training.
Perplexity	PerplexityBot (search results); Perplexity-User (user-requested fetches)	Perplexity says these crawlers are not used to train foundation models	Blocking PerplexityBot reduces your presence in Perplexity results.
Google	Googlebot (Search, including AI Overviews and AI Mode)	Google-Extended (a robots.txt token for Gemini training and grounding)	Google states Google-Extended “does not impact a site’s inclusion in Google Search” and is not a ranking signal.

The practical rule: you can allow search crawlers and still refuse training. OpenAI’s documentation gives exactly that example. Also note that user-triggered fetchers (ChatGPT-User, Perplexity-User) may not follow robots.txt, because a person asked for the page.

A check a professional runs: open yoursite.pt/robots.txt, then also check the firewall or CDN bot settings. A site can have a perfect robots.txt and still return a block page to AI crawlers because a “block AI bots” toggle was switched on at the hosting level. Fetch the homepage with each crawler’s user agent and confirm you get a normal page, not a challenge.

Step 3: Google’s AI features follow normal Search rules

Google is unusually explicit here. Its documentation says there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”. To be shown as a supporting link, a page must be indexed and eligible to be shown in Search with a snippet. The controls are the normal ones: robots.txt for Googlebot, and nosnippet, max-snippet or noindex to limit what is shown.

So “AI Overview optimisation” for a local business is mostly ordinary good SEO. And for local money searches, AI Overviews are still rare. In our study of 138 “trade + city” searches in Flanders in October 2026, an AI Overview appeared on 4. The 3-pack appeared on 135. We have not measured Lisbon yet, but the lesson is the same: for a local business, Google’s AI visibility today runs mainly through your Maps listing. See what the Maps top 3 have in common.

Step 4: give the AI a single source of truth

Assistants lift short, self-contained passages. Help them:

• An “about” page that states the facts plainly: legal and trading name, address, phone, services, service area, opening hours, year founded, who runs it. One sentence per fact. This is the page you want other sites, and the AI, to agree with.
• Identical name, address and phone everywhere: website footer, Google profile, Facebook, every directory. Same spelling, same phone format. Fix the old listings rather than adding new ones.
• Answer-first service pages: open with a sentence that answers the buyer’s question (for example: “We repair automatic gearboxes for all car brands in Barreiro and Moita, with a written quote before any work starts”). Then the detail. A passage that makes sense on its own can be quoted.
• Structured data (LocalBusiness or the most specific subtype) that repeats the same facts in machine-readable form.

Step 5: earn corroboration elsewhere

Your own website is one voice. The AI also reads your Google profile (rating, review count, categories), review platforms, directories that rank in your trade, and local press or association pages that mention you. In our Belgian data, directories held 54% of the organic top-10 positions for local searches, so they are among the pages an assistant is most likely to fetch.

The work is unglamorous: claim and complete each relevant profile, make it consistent, and keep getting real reviews from real customers. There is no shortcut that survives a second look.

How to measure it (and why no one can promise it)

AI answers change with the wording, the day, the user’s location and the model version. Nobody can guarantee you will be named. What you can do is measure honestly:

1. Write a fixed list of 10 to 20 questions your customers actually ask, in the language they use.
2. Ask them in the same assistants, in a clean session, with the same location, once a month.
3. Log who is named, in what order, and which sources are cited.
4. Change one thing at a time, and look at the trend over months, not a single answer.

If someone shows you one screenshot as proof of “AI ranking”, ask for the question list and the monthly log.

Next step

Want to know whether ChatGPT and Google can read your site, and who they recommend instead of you? Ask for a free diagnosis. We check your crawler access, your profile consistency and a short set of real customer questions, and tell you what we would fix first. For the website side, see the local website checklist.