AI Local SEO: How Local Businesses Show Up Correctly in AI Answers

When someone asks ChatGPT or Perplexity for "a dentist open Saturday near downtown", the answer comes from one of two places. Either the model answers from what it learned in training, or it searches the web and reads pages before it writes. As of October 2026, every major engine decides per query whether to search, so you never know in advance which path a given answer took.

It helps to split the process into four stages:

• Parametric: what the model already knows. OpenAI's model docs list a knowledge cutoff of February 16, 2026 for the GPT-5.6 family and August 31, 2025 for the Instant path ChatGPT also uses. Hours you changed after that date can't reach an answer given without search.
• Access: whether a crawler is allowed and able to fetch your pages.
• Retrieval: whether your page enters the model's context for this query.
• Generation: which retrieved pages the model actually uses and cites.

Engines also rewrite the question. OpenAI says ChatGPT search rewrites a prompt into one or more targeted queries, and Google may fan out into several related searches. That means you can't optimize for the exact words a customer types. Local answers can also vary by place and language, which is covered in how AI Overviews differ by location.

The practical point: you have real influence over access, some indirect influence over retrieval, and very little over the rest.

Most local SEO advice blurs these together. Separating them tells you where to spend a week and where to spend nothing.

The access row is where the strongest evidence sits, and its fixes work by removing blocks. Nothing you add to a page guarantees a mention. What you can do is make sure that when an engine looks, it finds correct facts it can read and quote.

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they can't read content that only appears after the page runs scripts. Googlebot renders JavaScript. Most AI search crawlers, as of October 2026, do not.

Say you run a dental clinic in Austin (a hypothetical). Your opening hours come from a booking widget. Your prices sit in a tab that loads on click. Your list of accepted insurers is an image. A person sees everything. OAI-SearchBot, PerplexityBot and Claude-SearchBot see a clinic name, a phone number and an empty box.

How to check

1. Open your contact or location page and view the page source (not the inspector).
2. Search the source for your Saturday hours, a price and your street address.
3. If any of them are missing, they arrive by JavaScript and need to move into the server-rendered HTML.

Site builders and headless setups are the usual cause. See whether a headless CMS hides your site from AI crawlers, and if you run WordPress, the WordPress settings that matter.

Write facts as quotable sentences

The one positive signal for on-page content is plain: a specific fact on a fetchable page gives the engine a sentence it can quote. Write "We're open Saturday 8am to 1pm" and "A new-patient exam costs $150" (a hypothetical figure) as text. Keep each fact identical everywhere it appears on your site.

Check you aren't hiding content by accident

Google documents noindex, nosnippet, max-snippet:0 and data-nosnippet as controls that keep content out of AI features. Bing applies data-nosnippet to its AI summaries too. Templates sometimes carry these from an old cookie banner or footer fix. Search your page source for them.

robots.txt works per bot, and each engine uses different bots. OpenAI's crawler page states that each setting is independent: GPTBot governs training, and OAI-SearchBot controls whether you appear in ChatGPT search. Blocking GPTBot doesn't remove you from ChatGPT search. Allowing it doesn't add you.

A minimal set of rules that allows the main AI search crawlers looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

Meta is the one operator that states allowing its crawler helps: it says allowing Meta-WebIndexer helps it cite and link to your content in Meta AI's responses.

Google is different. There is no robots.txt token for AI Overviews or AI Mode. Google says a page must be indexed and eligible to show with a snippet, and that there are no additional technical requirements. Googlebot access is the control.

The gate most sites never test

OpenAI's help centre says eligibility needs two things: allow OAI-SearchBot, and confirm your host or CDN allows traffic from its published IP addresses. A firewall can return 403 to a crawler that robots.txt welcomes. From September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, while Search crawlers stay allowed by default. A site can lose some AI access without anyone editing it.

The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

For local questions, engines often cite directories, review sites, maps listings and local news instead of the business itself. The evidence shows a strong association: when engines cite other sites, being named correctly on those sites is the route into the answer. It is not a promised effect, but it is the part of retrieval you can actually touch.

Work through it like an audit, not a submission spree:

• List the pages that come up when you search your category and town in Google and Bing.
• Open each one and compare your name, address, phone, hours, services and prices with your own site.
• Fix every mismatch at the source. An old Saturday schedule on a directory is a wrong fact waiting to be quoted.
• Retire duplicate listings from old addresses or previous business names.

Check Google and Bing indexing separately. In Otterly's 2026 experiment, Google deindexed two test sites while Bing kept them, and Microsoft Copilot, which grounds on Bing, kept citing them while Google's AI features almost never did. One index being healthy tells you nothing about the other.

The local business fact check for AI answers gives a field-by-field list. If you moved or renamed, also run the rebrand checklist for AI answers.

Do these in order. Each step depends on the one before it.

1. Write down your canonical facts. One document: legal and trading name, address, phone, hours by day, holiday hours, services, starting prices, insurers or payment options.
2. Make your own site match it. Every page, the footer, the contact page and any PDFs. Contradictions on your own site are the easiest wrong facts to fix.
3. Move those facts into server-rendered HTML. View the source and confirm each one is there as text.
4. Remove accidental suppression. Check for noindex, nosnippet, max-snippet:0 and data-nosnippet on pages that hold the facts.
5. Open robots.txt per bot. Allow the search crawlers you want. Decide on training bots separately.
6. Test the firewall. Request a page as each crawler and look for 403s or challenge pages.
7. Confirm indexing in both Google Search Console and Bing Webmaster Tools.
8. Align third-party listings with your canonical facts.
9. See what engines say now. The free checker on this site shows whether AI engines name you for up to three buyer questions. It is a snapshot, so repeat it after changes. For ongoing options, see affordable ways to monitor what AI chatbots say.

The classic basics still carry weight here. The SEO basics that matter for a small business site overlap heavily with this list.

• A blanket AI block in robots.txt. Blocking every AI user agent to stop training also removes you from ChatGPT search and Perplexity. Block GPTBot if you want, and allow OAI-SearchBot.
• Allowing GPTBot to "get found". It governs training only. It has no documented effect on ChatGPT search.
• Treating schema as a citation lever. Google says structured data isn't required for generative AI search, and Ahrefs' matched test found AI Overviews citations fell 4.6% after pages added schema. Existing markup does no harm. It just isn't the fix.
• Publishing an llms.txt and waiting. Google says Search ignores it, and no other engine documents reading it.
• Mass-producing location pages with AI. Google's spam policy names generating many pages without adding value as scaled content abuse. In Otterly's test, Google removed both 1,000-post AI sites with no manual action notice.
• Hours in images or widgets. If a crawler can't read the text, it can't quote it.
• Reading one blended score. Engines draw on different indexes. Check each one separately, or a strong result on one hides a gap on another.

Does LocalBusiness schema help me appear in AI answers? There is no evidence it does. Google says structured data isn't required for its generative AI features, and Ahrefs' matched-control study found adding schema produced no major uplift on any platform it tested. If you already have LocalBusiness markup, leave it, because Google says existing markup does no harm. Spend your time on rendered facts, crawler access and consistent listings instead.

Why does ChatGPT give my old opening hours? Two common causes. If the answer came without a web search, it reflects training data frozen at the model's cutoff, which for ChatGPT's Instant path is August 31, 2025. If it did search, it may have read an outdated directory or a page where your new hours load by JavaScript. Fix the listings, put hours in plain HTML, and check that OAI-SearchBot can reach your site.

Can I block AI training but still appear in AI search? Yes, on engines that separate the two. OpenAI states its settings are independent, so you can disallow GPTBot and allow OAI-SearchBot. Anthropic separates ClaudeBot from Claude-SearchBot. For Gemini Apps, Google-Extended is the control, and it doesn't affect Google Search. AI Overviews and AI Mode have no separate token: they follow Googlebot and normal index eligibility.

How often should a local business recheck its AI answers? Recheck after any change to hours, prices, address or name, and after site or CDN changes, since a firewall update can block crawlers silently. Outside those moments, a monthly look is a sensible rhythm for most small businesses. Engines change models and defaults often, so treat each check as a snapshot and compare engine by engine rather than chasing one number.