Find competitor keywords with AI: what a chatbot can and cannot tell you

Can you find competitor keywords with AI? Ask a chatbot what a rival ranks for and it will answer with confidence. Whether that answer is worth anything depends on one thing: did the model look at ranking data, or did it read the rival's pages and guess? Those are two different jobs, and only one of them tells you where a competitor actually shows up in Google.

What a model without data does

A language model with no search data, or with only a web browser, can do what a person does by hand. It can read a competitor's pages, pull out the phrases in titles and headings, notice the topics they write about, and suggest related searches. That is useful. It is also exactly the information you get from reading a sitemap or running Google Keyword Planner on a competitor URL, which Google describes as suggesting keywords related to the content of a site.

What it cannot do is tell you the position of a page or how many visits a keyword sends. Those numbers come from collecting Google results across a very large set of queries and recording which URL ranks where. A model that has not been given that data is inferring intent and presenting it as outcome. A competitor can publish pages for hundreds of phrases and rank for a fraction of them; a model reading the pages cannot tell which fraction.

What a model with a data tool does

The picture changes when the model can call a keyword data tool. Then the model handles the parts it is good at, such as choosing what to look up, filtering, grouping and explaining, while the tool supplies the rankings.

seogeoaeo.ai offers this through MCP and REST. Its Competitor Keywords tool takes a domain, or any page on it, and a country, and returns up to 100 keywords the whole domain ranks for in Google, ordered by estimated traffic. Each row has position, volume, difficulty, intent, cost per click, estimated visits and the ranking URL. It also separates out blog topics (informational searches with difficulty 50 or less, at least 10 searches a month, brand names excluded), keywords worth bidding on, and the pages bringing the site the most visits.

Connect Claude Code, Cursor, Codex or another MCP client to https://seogeoaeo.ai/api/mcp with a workspace API key and the model can call it by name, competitor-keywords. A run costs 35 credits, the same as from the browser. The call accepts maxCredits, so you can cap what any single run may cost, and an idempotencyKey so a retry does not charge twice. The details are in the machine-readable tool list.

A division of labour that works

A prompt along these lines works: "Run competitor-keywords on rival.example for the UK with maxCredits 35. Remove brand searches. Group what remains by ranking URL and list keywords where they rank 8 to 20 that match our product."

Where the numbers still need care

Adding a model does not make the data more exact. Volumes and visits are estimates from a search data provider, not Google figures, and two providers will disagree on the same site. Data is per country, so a model asked about a British rival in US data will report a smaller company than exists. The tool covers the whole domain rather than one folder, and it does not show which keywords a rival bids on in Google Ads, because Google does not publish those.

It is also a snapshot. seogeoaeo.ai does not track rankings over time; for a running history of your own positions, a rank tracker such as Semrush Position Tracking is the better tool. An agent can still store each month's result and compare.

Cost

Plans start at $29 a month for 2,000 credits, and every tool shows its price before it runs. An agent that first calls Competitor Finder (38 credits) to pick three rivals, then looks up each one, spends about 143 credits.

The takeaway

AI is good at the reading and sorting. It is not a source of ranking data on its own. Give it a tool that has that data, cap the spend, and keep the final call on which keywords to chase.