How to Get Your Business Recommended by ChatGPT, Gemini, Perplexity and Google AI

If you have ever typed your own category into ChatGPT — "best bookkeeper in Oshawa", "top HVAC company near me" — and watched three competitors come back while you did not appear, this article is the fix list.

Why AI recommendations are not the same as Google rankings

Assistants answer in two different ways, and the difference matters.

Some answers come from what the model already absorbed during training. Others come from a live web fetch at the moment you ask, which the assistant then summarises and cites. Business recommendation queries — anything with a city, a service category, or a word like "best" or "vs" — almost always trigger the live fetch.

That live layer runs on a different, much shorter list than Google's. Being on page one is table stakes; the assistant still has to choose you out of the handful of sources it retrieves. Industry measurements consistently show AI assistants recommending a far narrower set of local businesses than Google's local pack does, and each assistant leans on a different mix of sources, so a single-platform strategy travels badly.

Step 1: Let the AI crawlers in

Most sites that are invisible in AI answers are invisible for a boring reason — something is blocking the bots. Often it is not even robots.txt; it is a CDN or security plugin treating AI crawlers as scrapers.

There are two families of bot and you should treat them separately:

Training crawlers build the model's background knowledge: GPTBot, ClaudeBot, Google-Extended, CCBot, Applebot-Extended. Blocking these is a legitimate content-licensing decision.

Retrieval and user-action bots fetch pages at the moment someone asks a question: OAI-SearchBot and ChatGPT-User, Claude-SearchBot and Claude-User, PerplexityBot and Perplexity-User. Blocking any of these makes you ineligible for citation in that assistant's answers. There is no upside to blocking them if you want to be found.

A permissive baseline looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Perplexity-User
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

User-agent: Google-Extended
Allow: /

User-agent: *
Disallow: /account/
Disallow: /checkout/
Disallow: /cart/

Sitemap: https://example.com/sitemap.xml

Two things to verify after you edit it. First, check your CDN or WAF separately — robots.txt permission means nothing if Cloudflare is returning 403 to those user agents. Second, check your redirects. Retrieval bots are less patient than training crawlers, and a chain of redirects can be enough for the page to be dropped from an answer.

Step 2: Make your business facts identical everywhere

Assistants cross-check. When your hours say one thing on your site, another on Yelp, and a third on an old directory listing, the safe move for the model is to recommend a business whose details agree with themselves.

Gemini is grounded directly in Google Maps data, which makes your Google Business Profile the single highest-leverage asset for Gemini specifically. ChatGPT and Perplexity pull from a broader spread: Google Business Profile, Bing Places, Apple Maps, Yelp, LinkedIn, industry directories, and review platforms specific to your vertical.

Practical version:

• Write your name, address and phone in one exact format and use that format everywhere, down to the "Suite" versus "#".
• Claim and complete your profile on Google, Bing, Apple Maps and the two or three directories that actually matter in your industry.
• Kill or correct stale listings. Old addresses are worse than no listing.
• Treat reviews as a gate rather than a dial. Reporting from local-visibility studies suggests assistants effectively exclude businesses below roughly a 4.0 average rather than merely ranking them lower — and that response rate to reviews matters alongside the rating itself.

Step 3: Write pages that answer the question, not pages that rank for the keyword

A retrieval system is looking for a passage it can lift as an answer. Give it one.

The format that works is almost mechanical:

• Make the H2 the literal question a customer would ask, in their words.
• Put a direct 40 to 60 word answer immediately underneath it.
• Then expand with the detail, caveats and examples.

One page per real question beats one long page covering everything, because the retrieval unit is the passage and a heading that matches the question phrasing is one of the strongest selection signals.

The page types that get pulled into recommendation answers most often are the ones many businesses avoid writing: transparent pricing pages, honest "us vs them" comparisons, "who this is not for" sections, and specific service-area pages that say something real about the area rather than swapping the city name.

Include the boring specifics — numbers, dates, materials, turnaround times, named author, a visible last-updated date. Vague marketing copy gives a model nothing to cite.

Step 4: Get mentioned on sites you do not own

This is the step most businesses skip and it is often the deciding one. Assistants weigh how many independent credible sources say the same thing about you. Your own website is one source with an obvious bias.

What counts:

• Being included in someone else's "best X in [city]" roundup.
• Genuine participation in Reddit and niche forums, which several assistants pull from heavily.
• Local press, chamber of commerce pages, supplier and partner listings.
• Podcast appearances and YouTube, both of which are transcribed and indexed.
• Case studies published by your clients or vendors.

A short, unglamorous tactic that works: find the roundup articles that already rank for your target question, and email the author with a specific reason you belong on the list. Being added to three existing lists usually beats publishing a fourth one yourself.

Step 5: Add structured data

Schema does not force a citation, but it removes ambiguity about what you are.

At minimum: Organization or LocalBusiness with a complete address, opening hours and a sameAs array linking every profile you own. Add FAQPage markup on pages built around questions, Service or Product markup on your offer pages, and Article markup with a real author entity on your blog.

Step 6: Measure it, or you are guessing

Pick the twenty questions a real customer would actually type. Run all twenty through ChatGPT, Gemini, Perplexity and Google AI Overviews once a month. Record three things each time: did you appear, who else appeared, and which URLs were cited.

Then instrument the other side. In GA4, segment referral traffic from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. The volume will look small next to organic search — the intent behind it usually is not.

SEO4i does the first part of this for you: every scan includes an AI Search Readiness score covering crawler access, answer-first structure and entity clarity, and the AEO report tracks which of your live search queries trigger AI Overviews and whether your domain is among the cited sources. Run a free scan to see where you currently stand, compare the plans, or create an account to track it every month.

What does not work

Stuffing keywords into headings. Retrieval matches meaning, not repetition, and awkward phrasing reads badly to the humans who arrive afterwards.

Bulk-generated articles. Volume without substance gives assistants nothing distinctive to cite, and it dilutes the pages that might have been cited.

Treating llms.txt as a solution. It costs nothing to add a file pointing agents at your best pages, and it may well become a standard. Right now, adoption is not broad enough to treat it as a strategy on its own.

How long this takes

Perplexity re-crawls continuously, so a well-structured page on an already-indexed domain can appear in its answers within days. A brand-new domain takes longer to enter the pool at all.

ChatGPT is slower and lumpier, because part of its impression of your brand is baked into training rather than fetched live. The live-search layer responds faster than the model's background knowledge does, which is exactly why the third-party mention work in Step 4 pays off before anything else.

Expect the directory and review corrections to move first, the content structure changes to land over a few weeks, and the citation-density work to compound over months.

Common questions

Does blocking GPTBot stop me appearing in ChatGPT? Not entirely. GPTBot is the training crawler. The bot that fetches pages when someone asks ChatGPT a live question is OAI-SearchBot, with ChatGPT-User handling explicit link visits. You can block training while staying eligible for citation, but blocking the search bots removes you from answers.

Do I need a separate strategy for each assistant? No, but you do need broad coverage. The assistants cite overlapping-but-different source mixes, so accurate listings plus third-party mentions across several platform types travel better than optimising for one engine.

Will good Google rankings carry me into AI answers? Partly. Strong rankings help, but a large share of what assistants cite does not sit in Google's top ten for the same query. Treat AI visibility as a related but separate channel.

How often should I re-check? Monthly is enough for most businesses. Answers vary between runs even for the same question, so look at the trend across several checks rather than reacting to a single result.