ChatGPT recommends local businesses it can find, read, and corroborate: companies with machine-readable websites, an unambiguous identity, a healthy review footprint, and details that match everywhere the model looks. There is no submission form and no paid placement — the way in is to be the business the engine keeps encountering, described the same way, across every source it checks.
Where do ChatGPT's recommendations come from?
Two places. The first is training data: the model's compressed memory of the open web. If your business has been written about, listed, and reviewed for years, some of that likely made it in — which is why long-established businesses sometimes get named with no recent effort at all. The second source, and for local queries usually the decisive one, is live retrieval. Ask ChatGPT for a plumber in your town and it will typically run a web search behind the scenes, read a handful of pages, and compose its recommendation from what it just read. That is good news: the contest is happening on the open web you can still edit today, not inside a training run that ended months ago.
Crawler access follows from that. OpenAI fetches content with identifiable crawlers — GPTBot for training, OAI-SearchBot for live search. If your robots.txt blocks them, you have opted out of being read. That is a defensible choice for a publisher protecting its archive; for a local business hoping to be recommended, it is locking the front door and wondering why nobody visits.
What signals appear to matter?
OpenAI does not publish a ranking algorithm, so anyone claiming certainty is selling something. But watch what assistants actually recommend across many markets and a consistent shortlist emerges.
A site a machine can read quickly
Retrieval is impatient. Clean HTML, fast loads, and pages that state plainly what you do and where you do it give the model something to work with. Pages that answer common questions directly — the discipline of answer engine optimization — are the easiest for an assistant to trust and reuse.
Entity clarity
The model has to resolve you into a distinct thing: one business, one name, specific services, a defined service area. Vague brand names, mismatched naming across pages, or a site that never quite says what you sell all make you harder to resolve — and a business the model cannot confidently identify is a business it quietly skips. Schema markup and a plainspoken About page help more than they get credit for.
Reviews and reputation
Assistants lean on corroboration. Review volume, recency, and what reviewers actually describe — across Google, Yelp, and industry-specific sites — give the model third-party evidence that you are real and good at the work. So do mentions in local publications and roundups. One glowing source is a claim; several agreeing sources are a fact, as far as a language model is concerned.
Consistency across sources
When your name, address, phone, and service list match across your site, your profiles, and the directories, the model's sources agree — and agreement is what makes you a safe name to say. Conflicting hours or a stale address make you a risk. This is classic local SEO hygiene doing double duty in a new arena.
What should a local business actually do?
- Check robots.txt and confirm you are not blocking GPTBot or OAI-SearchBot.
- Publish pages that answer the questions people actually ask assistants — cost, process, "who's the best X for Y" — directly and early on the page.
- Tighten entity signals: consistent naming, LocalBusiness schema, an About page that says the obvious things out loud.
- Keep reviews flowing and respond to them. The text of reviews is source material, not just a star count.
- Fix disagreements between your site and your directory listings — every mismatch is a reason for the model to hedge.
How do you know if it's working?
Ask the assistant — repeatedly, in different phrasings, over time. One-off spot checks mislead, because answers genuinely vary between runs and between wordings. What matters is the trend: are you named more often this month than last, and for which questions? This is the tedious part Speak Local automates — putting your market's questions to the engines on a schedule and recording who gets named — but whether you use a platform or a stubborn spreadsheet, measure it. Being recommended by a machine is now a business outcome. It deserves a number.
