How to get recommended by ChatGPT, Claude and AI agents

That matters because more buyers now ask an assistant which tool to use, and developers let a coding agent pick and install one. This guide covers what each assistant documents about where it looks, the five things you can influence, what to avoid, and how to check whether assistants mention you.

How do AI assistants decide which tools to recommend?

AI assistants recommend tools from two sources: what the model learned in training, and what a live search or page fetch returns when someone asks. Neither OpenAI, Anthropic, Perplexity nor Microsoft publishes how its assistant ranks products, so any article claiming to know "the ChatGPT recommendation algorithm" is guessing.

What is documented is the retrieval step. Microsoft, for example, explains that Copilot turns a prompt into a short search query and "sends to the Bing search service asking for more information," then uses the results to compose its answer (Microsoft Learn). Google says its AI features need nothing special: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" (Google Search Central).

So "how to show up in ChatGPT" breaks into two jobs. First, be one of the pages the assistant retrieves. Second, be the option its answer names. You can work on both, but only the first is mostly in your hands. For the retrieval mechanics engine by engine, see how AI search chooses its sources.

Can each AI assistant reach your pages?

Each assistant searches a different index, and a page missing from that index can't be recommended from live search. As of October 2026, this is what each operator documents:

The training crawlers (GPTBot, ClaudeBot) are separate settings; blocking them doesn't remove you from these search answers. Our guides to OAI-SearchBot and ClaudeBot and Anthropic's other bots cover the rules, and the AI crawlers guide compares every major bot. For Copilot, verify your site in Bing Webmaster Tools and send changed URLs with IndexNow (an open "this page changed" notice that Bing and other participating engines accept; Google doesn't use it). The engine guides for ChatGPT SEO, Claude and Perplexity go deeper on each.

For Google's AI Overviews and AI Mode, the first job is being in Google's index. Check which of your pages Google has indexed, free.

Do comparison pages help AI assistants recommend you?

Comparison pages help because people ask assistants comparison questions, such as "X vs Y" or "which tool should I use for Z", and a fair, sourced comparison answers them directly. A page that only praises you gives an assistant little it can check against other sources.

What makes a comparison usable, and honest:

• Facts from each vendor's own pages, linked and dated "as of <month year>", in a table an assistant can read row by row.
• A real "when to choose them" section. If a competitor suits some buyers better, say so.
• A disclosure line when you compare your own product. In the US, the FTC's rule on reviews prohibits "misrepresenting that a website or entity it controls provides independent reviews or opinions about a category of products or services that includes its own products or services" (FTC, August 14, 2024). Our own tool listings open with "Disclosure: we make FreshlyCited."

What docs can a coding agent read and act on?

A coding agent can recommend a tool confidently when it can read the setup docs and finish the install without help. Agents read docs much as search crawlers do, so the basics carry over: real text in the HTML, one topic per page, the key facts near the top.

Claude Code's documentation shows why. Its WebFetch tool converts a page's HTML to Markdown and reads "up to 100,000 characters of a page's content per call." It then runs an extraction prompt over that content, which Anthropic calls "lossy by design" (Claude Code docs). A setup page that buries the install command under marketing copy risks losing it in that step.

Practical habits for agent-readable docs:

1. One-command install and a copy-paste config block, in fenced code.
2. A plain Markdown copy of each docs page, if your docs platform can produce one.
3. An llms.txt file listing your key docs. Its author describes it as "a proposal to standardise on using an /llms.txt file to provide information to help agents use a website" (llmstxt.org). It's a convenience for agents, not a ranking signal; Google says it doesn't need "AI text files" (see what llms.txt does and doesn't do).
4. An AGENTS.md snippet your users can paste into their own repos. AGENTS.md is "a README for agents", "used by over 60k open-source projects" and stewarded by the Agentic AI Foundation under the Linux Foundation (agents.md). Ours is at the FreshlyCited AGENTS.md snippet.

Should you publish an MCP server and list it in the registry?

Publish an MCP server (Model Context Protocol: a standard way for AI assistants such as Claude or Cursor to use a tool directly) if your product is something an agent could operate, not just describe. An agent that can call your tool can complete the task it was asked to do, instead of only naming you as an option.

The official MCP Registry is where public servers are listed. As of October 2026, its documentation says:

• "The MCP Registry is currently in preview. Breaking changes or data resets may occur before general availability."
• It is "the official centralized metadata repository for publicly accessible MCP servers." Private servers aren't supported.
• Names use a reverse-DNS format such as io.github.username/server, and publishers verify ownership "through GitHub, DNS, or HTTP challenges."
• It is "intended to be consumed primarily by downstream aggregators, such as MCP server marketplaces," which pull its metadata on a regular schedule.

Source: MCP Registry docs. You describe the server in a server.json file and publish it with the mcp-publisher CLI (init, login, publish) (registry quickstart). The registry also has a public API. We searched it on October 17, 2026:

curl -s "https://registry.modelcontextprotocol.io/v0/servers?search=indexnow&limit=10" \
  | jq '.metadata.count, [.servers[].server.name]'

3
[
  "app.openkrill/indexnow",
  "com.aiwebsitepipeline/indexnow-toolkit",
  "io.github.sharozdawa/indexnow"
]

Each entry also carried description, version, remotes and repository fields, so write a description that says plainly what the server does. Some clients have their own directories as well. Claude Code's docs point users to "reviewed connectors in the Anthropic Directory" and add a remote server with claude mcp add --transport http <name> <url> (Claude Code docs). Our comparison of SEO MCP servers shows how listings in one category differ.

Why do third-party mentions matter for AI recommendations?

Mentions on other sites matter because an assistant that reads several sources can see whether others describe you the same way you do. In a study of 75,000 brands published May 26, 2025, Ahrefs found that "Web mentions (0.664) correlate much more strongly than backlinks (0.218)" with brand visibility in Google's AI Overviews, and its author adds "correlation ≠ causation" (Ahrefs). The study covers AI Overviews only, not ChatGPT, Claude or Perplexity.

Mentions you can earn honestly:

• Accurate listings in the directories, marketplaces and registries your buyers and their agents use.
• Real customer reviews on established review sites, requested without conditions on what customers say.
• Helpful answers in communities, with a clear disclosure when you mention your own product.
• The same one-line description of what you do, everywhere you're listed.

A mention only helps if the page it sits on can be found. If a review page or directory listing isn't in an assistant's search index, live search can't retrieve it.

What should you not do to get recommended by AI?

Don't try to manipulate the model. The tactics that fail with search engines fail here too, and some carry legal risk:

How do you check whether AI assistants recommend you?

You check by running a fixed set of buyer questions across assistants on a schedule, because answers vary by run, wording and time. Record for each prompt whether you were named, whether you were linked, and which page was cited, then judge the trend over several weeks.

Our AI visibility tracking guide sets up a weekly routine. Trackers such as Otterly (as of October 2026) automate the prompt runs across ChatGPT, Perplexity, Copilot and Google's AI features, and our page on using FreshlyCited with Otterly shows how the citation data and index status fit together. For Copilot, the AI Performance report in Bing Webmaster Tools shows "how often your content is cited" across Copilot, Bing's AI summaries and select partners (Bing Webmaster Blog). When a page you expected to see never appears, check access and indexing before rewriting it.

What can nobody promise about AI recommendations?

Nobody can promise that ChatGPT, Claude, Perplexity or a coding agent will recommend your product, and none of the operators documents a way to buy or request a place in an assistant's answers themselves. Each assistant decides on its own which sources to read and which options to name, and none publishes its criteria.

What you control is the evidence it finds: pages it can reach, comparisons that are fair, docs an agent can follow, a listing for your MCP server, and an accurate description of you on other sites. Our AI search optimization guide covers the wider picture.

Where does FreshlyCited fit?

FreshlyCited covers one step, Google indexing, and not the recommendation. It checks whether Google has indexed your pages and the third-party pages that mention you, asks Google to look at the ones it hasn't, and re-checks on days 1, 3 and 7; being indexed is what makes a page eligible for Google Search, AI Overviews and AI Mode. For sites you verify with an IndexNow key, it also sends changes to Bing. "ChatGPT, Perplexity and other assistants use their own crawlers and search partners. We don't submit to them, and we don't claim to." Its MCP server lets Claude, Cursor and other MCP clients run those checks directly; see the API and MCP server (MCP is on the Agency Pro and Agency Scale plans). It can't make any assistant recommend you, and neither can anyone else. The free plan gives you 10 free links to try, then 100 free checks a month, with no card. Check my pages free.