Guide · AI search
How to Get Recommended by ChatGPT and Perplexity
When someone asks an AI assistant to find them a plumber, a dentist or a place to eat, the assistant answers from pages it can read and sources it considers credible. Getting recommended is not a setting you switch on. It comes down to three things: letting the crawlers in, structuring your pages so an answer can be lifted out of them cleanly, and being described consistently enough across the web that the system is confident you exist.
By Carlos Boluda · Updated · 8 min read
Being cited is a different target from being ranked
Traditional search sends someone to your page. An AI answer usually does not — it reads your page, extracts what it needs, and tells the person directly, sometimes with a link and sometimes without one.
That changes what you optimise for. Keyword density is close to irrelevant. What matters is whether a specific, self-contained answer exists on your page in a form that can be quoted. A paragraph that only makes sense after reading the three before it cannot be extracted, so it will not be.
The practical shift is from thinking about pages to thinking about paragraphs. Every section should survive being pulled out and quoted on its own.
Step zero: check you are not blocking the crawlers
AI systems use their own crawlers, and a robots.txt written years ago may be blocking them without anyone noticing. The names to check for are GPTBot, PerplexityBot, ClaudeBot, Google-Extended and Applebot-Extended.
A wildcard rule allowing everything covers them all. An explicit block on any of those names means zero visibility on that platform, no matter how good the rest of your site is. This is the single highest-priority thing to verify, and it takes thirty seconds: open your-domain.com/robots.txt and read it.
The second technical check is whether your content exists without JavaScript. Crawlers frequently do not run it. If your page is empty when scripts are disabled, everything below this line is wasted effort.
Structure beats length
The formats that get extracted are consistent and boring: a direct definition in the opening, numbered steps for anything procedural, tables for comparisons, and explicit question-and-answer pairs.
Prose is the weakest format for this. A comparison written as two paragraphs of narrative will lose to the same comparison written as a four-row table, because the table can be lifted whole and the paragraphs cannot.
This is also why burying the answer works against you. If the question is "how much does X cost", the answer belongs in the first two sentences under a heading that asks that question, not in a conclusion after eight hundred words of context.
Why unattributed numbers hurt you
A statistic with no source cannot be verified, and systems that are being judged on accuracy deprioritise what they cannot check. "Most businesses see a 40% increase" is worth less than nothing — it is a reason to trust the rest of the page less.
The fix is mechanical: name the organisation and the year, or delete the number. A claim you can attribute becomes a quotable unit. A claim you cannot attribute is a liability on your own page.
The same applies to vague authority. "Experts recommend" and "studies show" are empty. Name the expert, name the study, or cut the sentence.
Consistency is what makes you an entity
Before a system can recommend you, it has to be confident you are a real, specific business. That confidence is built from agreement across sources: your site, your Google Business Profile, directories, social profiles, anywhere you are mentioned.
Contradictions weaken it. A different phone number on an old directory listing, a business name written three ways, an address that changed two years ago — each one is a small reason to be less sure, and being less sure means being less likely to name you.
Structured data helps here, and this is where many sites quietly undermine themselves. Schema markup is read directly by machines, so anything inaccurate in it is taken at face value. A placeholder phone number left in schema is worse than no phone number at all.
How to check whether any of this is working
There is no dashboard for AI citations. The only reliable method is manual and it takes twenty minutes: write down ten questions a customer would ask, put each one to ChatGPT, Perplexity and Google's AI answers, and record who gets named.
Pay attention to the format of what wins, not just the name. If the cited sources are using tables and question headings while you are using prose, you have your answer without needing any further analysis.
Repeat it monthly. It is crude, it is the only thing available, and it is more informative than any tool currently claiming to measure this.
What to change, and what it affects
| Change | Why it matters | Effort |
|---|---|---|
| Allow AI crawlers in robots.txt | A block means zero visibility on that platform | 5 min |
| Content present without JavaScript | Crawlers often do not execute it | Varies |
| Definition in the first 200 words | Gives the system something to lift | 15 min/page |
| Question-shaped headings | Matches how people phrase things to an assistant | 20 min/page |
| FAQ block with FAQPage schema | Highest-impact single change for most pages | 45 min/page |
| Tables instead of narrative comparisons | Tables get extracted, prose usually does not | 30 min/page |
| Sources named on every statistic | Unverifiable numbers are deprioritised | 10 min/page |
| Named author with a real bio page | Attribution is a credibility signal | 1 h, once |
| Consistent details everywhere | Builds confidence that you are a real entity | 2 h, once |
| Accurate structured data | Read literally. Placeholders do real damage | 30 min, once |
How to make a business website citable
- 01
Open your robots.txt and read it
Check that GPTBot, PerplexityBot, ClaudeBot and Google-Extended are not blocked. A wildcard allow covers them. This is thirty seconds and nothing else matters if it is wrong.
- 02
Disable JavaScript and reload your own site
If the page is blank or the text is gone, crawlers see the same thing. Fix this before writing anything new.
- 03
Put a direct answer in the first two sentences of every page
No warm-up, no scene-setting. State what the thing is and who it is for, then expand afterwards.
- 04
Rewrite headings as the questions people actually ask
"Pricing" becomes "How much does it cost?". The heading should match the words someone would say out loud to an assistant.
- 05
Add an FAQ block and mark it up with FAQPage schema
Five or six real questions, each answered in two or three sentences directly under it. This is the highest-impact single change on most pages.
- 06
Convert your comparisons into tables
Anything with two or more options being weighed against each other. Clean headers, one row per option.
- 07
Attribute or delete every number on your site
Name the organisation and the year, or cut it. An unsourced statistic makes the whole page less trustworthy to a system checking its work.
- 08
Audit your structured data for placeholders
Open your page source and read the JSON-LD. Fake phone numbers, template addresses and leftover sample values are read literally and taken at face value.
- 09
Run the ten-question test, then repeat it monthly
Ask the same ten customer questions on ChatGPT, Perplexity and Google's AI answers. Record who is named and in what format. It is manual and it is the only thing that works.
Frequently asked questions
How do I get my business recommended by ChatGPT?
Make sure your site is not blocking AI crawlers in robots.txt, put direct answers in the opening lines of each page, add an FAQ section with FAQPage schema, and keep your business details identical everywhere they appear online. There is no submission process or paid placement — recommendation comes from being readable and consistent.
Is AI SEO different from normal SEO?
It overlaps substantially but the target differs. Normal SEO aims to rank a page so someone clicks it. AI optimisation aims to have a specific passage extracted and quoted, so the unit of work is the paragraph or section rather than the page. Technical fundamentals and credibility matter to both.
Can I pay to appear in AI answers?
Not for the organic recommendations these assistants give. There is no advertising product that places a business inside an answer, and anyone selling guaranteed inclusion is selling something they do not control.
Which AI crawlers should I allow?
GPTBot for ChatGPT, PerplexityBot for Perplexity, ClaudeBot for Claude, Google-Extended for Google's AI answers, and Applebot-Extended for Apple. A wildcard rule allowing all user agents covers every one of them.
Does blocking AI crawlers protect my content?
It removes you from citation on that platform, and in practice the same crawler usually handles both training and answering. There is currently no reliable way to allow one and block the other, so blocking is a decision to be invisible there.
How do I know if an AI is recommending my business?
By asking it. No standard reporting exists, so the working method is to test ten customer questions across the main assistants every month and record who gets named. Google Search Console also reports impressions from AI answers separately, which is the one automated signal available.