How to show up in ChatGPT, Claude and Perplexity
The three assistants retrieve differently, so the same page performs differently in each. What each one favours, and the work that pays across all three.
The three assistants people actually use retrieve differently. The work overlaps a lot, and where it does not, the differences are worth knowing.
The one action most teams have not done
Verify your site in Bing Webmaster Tools and submit the sitemap. ChatGPT search and Microsoft Copilot both retrieve through the Bing index. It is free, it takes about twenty minutes, and most teams have never done it because Bing has not mattered for a decade.
Google Search Console does not help here. It is a different index.
ChatGPT
How it retrieves. ChatGPT search resolves through the Bing index. OAI-SearchBot indexes pages; ChatGPT-User fetches a page when a live question needs it. GPTBot is the training crawler and a separate decision.
What to do
- Bing Webmaster Tools: verify, submit the sitemap, resubmit after a content batch.
- Allow
OAI-SearchBotandChatGPT-Userinrobots.txt. Blocking those removes you from answers now; blockingGPTBotaffects future training instead. - Keep the content in the HTML source. Check with
curl. - Answer-first paragraphs of 40 to 60 words under question headings.
What it favours. Established, corroborated sources. If three neutral sites describe your product the same way you do, that consistency carries more weight than saying it louder on your own site.
Perplexity
How it retrieves. Its own index plus live fetching, through PerplexityBot and Perplexity-User.
What to do
- Allow both agents.
- Be specific. Perplexity rewards pages that answer the exact question with numbers rather than pages that survey a topic.
- Put a visible date on the page. It leans on recency noticeably.
- Use tables. They survive extraction and get reproduced.
What it favours. Perplexity cites heavily and visibly, which makes it the best of the three for finding out which of your pages is actually working. Look at what it cites for your category's questions, and write more of that.
Claude
How it retrieves. Through Claude-SearchBot for indexing and Claude-User for live fetches. ClaudeBot is the training crawler.
What to do
- Allow
Claude-SearchBotandClaude-User. - Favour completeness over length. A page that answers the question and stops does better than a page that pads.
- Structure matters: headings that are questions, tables for anything comparable, code that runs.
What it favours. Clarity. In our own experience, a page that states a limit plainly, including where the product is a bad fit, tends to get used rather than skipped.
The work that pays across all three
Everything below helps in every assistant, and most of it takes an afternoon.
| Work | Why |
|---|---|
| Answer-first paragraph, 40 to 60 words, under a question heading | The shape extraction looks for |
| Content in the HTML source, not injected by JavaScript | Many crawlers do not execute it |
| Specific numbers with dates | Models weight specificity, and a dated fact is quotable |
| Tables for anything comparable | Extracted reliably, reproduced whole |
| A visible published and updated date | Recency counts more than it did in classic search |
| Third party sources saying the same thing | Consensus outweighs self-description |
llms.txt and a Markdown twin of each key page | Cheap, and almost nobody does it |
AI crawlers allowed by name in robots.txt | Removes an own goal |
How to tell whether it worked
Do not rely on impressions. Ask the assistants your category's real questions, monthly, and record the answers. That is what the visibility tools automate, and you can do a useful version of it by hand in an hour with a fixed list of twenty questions.
Watch which of your pages get cited, not just whether you are mentioned. The cited page tells you what to write more of.
The limit worth stating
All of this puts you in front of a person reading an answer.
If you sell a developer tool, there is a second surface where nobody reads anything: a coding agent inside a repository, choosing and installing. In our measurements Claude Code ran a web search in only 1.6% of decision runs, so for those sessions none of the work on this page was in the room.
Do this work, because it is cheap and it compounds. Then measure install share separately.
Common questions
How do I get my site mentioned in ChatGPT?
Be in the Bing index, because ChatGPT search retrieves through it. Verify the site in Bing Webmaster Tools, submit the sitemap, allow OAI-SearchBot and ChatGPT-User, and publish pages that answer specific questions directly.
How do I get cited by Perplexity?
Perplexity cites heavily and visibly, and it favours pages that answer the exact question with specifics. Allow PerplexityBot and Perplexity-User, put a direct answer under a question heading, and include numbers with dates.
How does Claude retrieve pages?
Through its own search and fetch agents, Claude-SearchBot and Claude-User. Allow both. Claude tends to reward pages that are clear and complete rather than pages that are long.
What single action helps most across all three?
Bing Webmaster Tools. ChatGPT and Copilot both retrieve through the Bing index, and most teams have never verified their site there. It is free and takes about twenty minutes.
Where this comes from
Armature ran 5,292 judged sessions with Claude Code, Codex and Cursor inside 51 realistic codebases, and published every run. The numbers on this page come from that work.