How do AI assistants decide which companies to recommend?
They search the web for the question, read a handful of pages and name the companies those pages support.
ChatGPT, Claude and Google's AI Overview all answer buying questions by retrieving pages and summarising them. Google says its AI features are rooted in its core Search ranking and quality systems. ChatGPT and Claude run their own searches with their own crawlers. So the work has two parts: your own pages have to be readable and clearly answer the question, and other sites have to say good things about you, because an assistant names companies that several sources agree on.
Step 1: What do the engines say about you today?
Ask them the questions your customers ask, several times, and write down who is named.
- Write 10 to 20 questions the way a buyer would type them: "best accounting software for a small agency", "which running shoes are good for flat feet", not your company name.
- Ask each one in ChatGPT, Claude, Gemini and Google. Answers change from one run to the next, so ask each more than once and count how often you are named.
- Note who is named instead of you, and which pages the answers cite. Those pages are your map.
This is what the free Kortex audit does for you: 20 questions to ChatGPT, Gemini and Claude, Google's page one and AI Overview, and the pages cited, in about four minutes.
Step 2: Can the AI search crawlers read your site?
Check your robots.txt and your firewall, because a blocked search crawler means your pages cannot be cited.
Each company runs separate crawlers for search and for model training, and blocking one does not block the other.
- ChatGPT: OAI-SearchBot decides whether your site can appear in ChatGPT's search answers. GPTBot is the training crawler, and ChatGPT-User fetches a page when a user asks.
- Claude: Claude-SearchBot indexes pages for Claude's search; Anthropic says blocking it may reduce your visibility in Claude's answers. ClaudeBot is for training and Claude-User fetches on request.
- Google: AI Overviews use the normal Google index. A page has to be indexed and eligible to show with a snippet.
If you want to be recommended, allow the search crawlers even if you choose to block training. Then check that your CDN or bot protection is not quietly returning errors to them.
Step 3: Are your words in the page before JavaScript runs?
Make sure each important page serves its text in the HTML, because many AI crawlers do not run JavaScript.
Open your page, view the source (not the inspector) and search for a sentence from the page. If the text is missing and only appears after scripts load, an assistant may see an empty page. Server rendering or pre-rendering fixes this. This site is pre-rendered for that reason.
Step 4: Does your site answer the questions directly?
Give each important question a page or a section whose first sentence is the answer.
- Use the buyer's question as the heading, and answer it in the first one or two sentences.
- Say plainly what you sell, who it is for, what it costs and how it compares. If your price is behind a "contact us" button, an assistant cannot quote it.
- Write comparisons and "best for" pages only where you can say something true and specific. Thin pages written only for search do not help, and Google treats mass-produced pages as spam.
Step 5: Do you describe yourself the same way everywhere?
Use the same one-line description of your company on your site, your profiles and directory listings.
Assistants look for agreement between sources. If your site says one thing, your LinkedIn page another and an old directory listing a third, the answer will be vague or wrong. Pick one sentence that says what you do and for whom, and use it everywhere.
Step 6: Are you on the pages the answers cite?
Get mentioned on the sites that already show up in the answers for your questions.
The pages cited in Step 1 are usually review sites, industry roundups, community threads and comparison articles. Ask to be included where you genuinely belong, collect reviews from real customers, and take part honestly in the communities your buyers use. Never post fake reviews or fake accounts.
Step 7: Add structured data and an llms.txt
Both are small, cheap to add, and help some engines read you, but neither is required.
- Structured data (JSON-LD) states who you are and what you sell in a form machines read. Google says no special schema is needed for its AI features, and that structured data should match what the page shows.
- An llms.txt file is a short Markdown summary of your site for AI tools. See what llms.txt is and whether you need one.
How long does it take to start showing up?
Weeks to months, and it varies by engine, so check again on a schedule rather than once.
Crawlers have to revisit your pages, and the third-party pages that mention you have to be found too. OpenAI notes that robots.txt changes for its search crawler can take about 24 hours to take effect; content changes take longer to show up in answers. Run the same questions every week or two and watch the rate at which you are named, not a single answer.
How Kortex helps
The Kortex AI visibility audit runs Step 1 and checks Steps 2, 3 and 7 for you, and writes four fixes: an article, an FAQ, an llms.txt file and structured data. On the Growth plan Kortex checks again every week and emails you on Monday with what changed.