Check what ChatGPT, Claude, Gemini and Perplexity answer about your category

Why check the answers, not only the rankings

A buyer who asks an assistant for tools in your category gets a short list, not ten blue links. Whether you are on it, who is, and which pages the assistant leaned on are three separate facts, and a rank tracker shows none of them. Search answers the question what do people search for, and what do ai assistants answer? It reads Google Trends, ChatGPT, Claude, Gemini and Perplexity.

1. Connect TrendsFast to Claude

Create an API key in the dashboard under Add to your agent, set it as TRENDSFAST_API_KEY in your shell, and add the server to Claude Code in a folder of its own:

mkdir ai-answers && cd ai-answers
claude mcp add --transport http trendsfast https://trendsfast.com/api/mcp/v3 \
  --header "Authorization: Bearer $TRENDSFAST_API_KEY"

The Claude apps can use the custom connector instead, where your Add to your agent page offers it. Your project’s confirmed context decides the category and the competitors the answers are checked against, so name your competitors there before the first call.

2. Ask for a table, then for one page

Ask Claude to tabulate before it advises. Each row then comes from one answer the call returned, and the advice at the end can only point at what the table shows.

Use TrendsFast. First read my project context and my credit balance (both free).
Then run one Search call (trendsfast_niche_trends with product "search"
and my context_id), with max_credits 35 and a request_id you keep.
From data.ai_answers, make one table with a row per assistant:
the assistant and model, whether it names my brand (mentions_you),
which of my competitors it names, the other brands it names, and the domains it cites.
Then tell me:
1. which brands two or more assistants name, and which only one names;
2. which cited domains I could realistically appear on (a list, a review, a forum);
3. what the answers get wrong or leave out about my category, quoting the excerpt;
4. one page or section I should write that states those facts plainly.
Read the excerpts to see which question was asked; if it is not the question my
buyers ask, say so before drawing conclusions.
Do not call any search keyword rising unless the result measured it.

3. What it costs

• One Search call costs 35 credits and collects fresh before it answers. max_credits is a ceiling: a call whose quote is above it is refused with CREDIT_QUOTE_EXCEEDS_LIMIT before anything is charged.
• Keep the request_id. Sending the same one again replays the saved result instead of paying twice, and trendsfast_read_result rereads it for free.
• Run it once a week or after you publish: the answers change, and one reading is a snapshot, not a position.

An example: TrendsFast’s own category

We ran this for TrendsFast itself on 2 November 2026, with TrendsFast’s confirmed context. What came back:

• All four assistants answered. None of the four named TrendsFast or its two named competitors.
• The answers named general tools instead: Claude listed ten names, among them Intercom, Canva and QuickBooks; Perplexity named Canva and Zapier.
• Only Perplexity cited sources: eight domains, from small-business review sites to a scraping vendor’s blog. ChatGPT, Claude and Gemini cited none.
• The excerpts showed a broad small-business question rather than one about trend data for agents, which is why the prompt above asks Claude to check the question first.

The useful output was the gap: no assistant connects this category with agent-callable market evidence yet, so the page worth writing is one that says plainly what that is.

Evidence and limitations

• Each assistant is asked once, with one question. Answers vary between runs and between users; treat one reading as a sample, not a share of voice.
• Names are matched as written. A brand named under another spelling may be missed, and nothing is attributed by similarity.
• Search interest from Google Trends is relative, not volume. A keyword is only rising when the result says it measured growth.
• TrendsFast does not change what any assistant says. It shows the answers; what you publish, you publish.

When this is not the right approach

• You need keyword volumes or rank tracking for many pages: use a search tool for that, and this check for the few questions that decide a purchase.
• Your buyers do not ask assistants about your category, for instance when they buy through a tender or a reseller.
• You want to know what your customers are stuck on: start with the Reddit complaints guide and come back for the answers.