Competitor keyword spy tools: what the data really is, and the workflow that turns it into a plan

A competitor keyword spy tool does not read your competitor's analytics. It builds an estimate: a very large index of search results, crawled continuously, tells the tool which pages appear for which queries, and panel or clickstream data turns that into an estimated share of clicks. So the export you get is a ranked list of good hypotheses about where a rival earns attention, not a ledger of their traffic. Used that way it is one of the most productive hours in search marketing. Used as fact it produces a plan built on numbers nobody can reconcile. Two sources are exact rather than estimated, and both are free: Google's Ads Transparency Center shows the ads an advertiser is actually running, and the auction insights report in Google Ads shows which advertisers you genuinely compete with in your own auctions.

Below: what each kind of data is built from, the free layer, how to read an export without fooling yourself, and a workflow that ends in a decision.

Where the data comes from

Source	What it tells you	What to remember
Crawled results index	Which queries a competitor's pages appear for, and roughly where	A position is a sample, taken on a date, from one location and device
Clickstream and panel data	Estimated visits and estimated share of clicks	Modelled from a panel, then scaled. Directionally useful, rarely precise
Keyword volume databases	Average monthly searches for a term	Averaged over months and often grouped into bands, so seasonal terms read low
Ad transparency and auction reports	Ads actually served, and advertisers you actually overlap with	Exact, published by the platform, and free

The main platforms in the first three rows are Semrush, Ahrefs, Moz, Similarweb, SpyFu and Serpstat. They differ in index size, in how often each country is refreshed, and in how they model clicks, which is why two tools give two numbers for the same competitor. Pick one as your system of record and use a second only to sanity check a surprise.

The free layer, which answers more than people expect

• Ads Transparency Center. Search an advertiser and see the ads they are running, with formats and regions. No account, and it is the platform's own record rather than an estimate.
• Auction insights. In your own Google Ads account, the report gives six statistics for search campaigns: impression share, overlap rate, outranking share, position above rate, top of page rate and absolute top of page rate. Overlap rate is the one people miss. It names who is in your auctions, which is often not who you assumed.
• Keyword Planner. The planner will generate keyword ideas from a website address, so you can point it at a competitor's domain or at one product page and read Google's own volume and competition figures for what it finds.
• Your own Search Console. Export the queries where you already appear in positions five to twenty. Those are queries a competitor is currently winning with you on the same page of results, which makes them the shortest route to a gain.

How to read a competitor keyword export

Four habits separate a useful read from a misleading one.

• Strip the brand terms first. A rival's own name and product names will usually top their keyword list and tell you nothing you can act on. Filter them out before you look at totals.
• Treat traffic columns as a ranking device. An estimated visits figure is volume multiplied by an assumed click through rate for the position the tool last saw. It is good for sorting rows and poor for forecasting. For the detail of why, see our note on how to read a competitor traffic estimate.
• Check the page, not just the term. Open the ranking URL. A competitor holding a query with a thin page is an opening. One holding it with a deep, updated, well linked page is a different project and should be priced as one.
• Look at what is missing. A tool only reports terms that exist in its own keyword database. Newer phrasings and genuinely niche questions often do not, which is why a short round of customer calls still finds queries no export contains.

From export to plan in six steps

• Pick three real competitors. Not the three biggest names in your sector. The three that show up when your buyers search, which auction insights and your own results pages will tell you.
• Export each one's ranking queries for the country you sell in, and keep the ranking URL column.
• Filter to commercial intent. Keep the queries that imply a purchase, a comparison or a problem you are paid to solve. Discard the informational tail unless it feeds a buying question.
• Group by the page that would answer it. Twenty queries that one page can satisfy are one job, not twenty. A clustering pass is where the plan appears. Our five step keyword research method sets out the grouping.
• Score each cluster on value to the business, the strength of what currently ranks, and whether you can produce something better. Three numbers, no model.
• Decide, and write the briefs. Five clusters with owners and dates beats a spreadsheet of nine hundred rows. Then run the same export quarterly to see which gaps a rival closed.

For the mechanics of comparing two keyword sets in one platform, our walkthrough of the Semrush keyword gap report goes step by step. For watching rivals continuously rather than in a one off study, see monitoring competitors in search.

Four errors worth avoiding

• Copying a competitor's whole keyword list, which copies their strategy including the parts that are not working.
• Comparing a tool's estimate of their traffic with your own analytics figure. One is modelled, the other is measured, and the comparison has no meaning.
• Chasing the highest volume term on the list when the pages ranking for it are all national brands with years of links.
• Running the export once. The value is in the quarterly difference, because that shows you what a rival decided to invest in.

Competitor keyword questions that decide how you use the data

Is using a competitor keyword tool legal?

Yes. These tools read public search results and licensed panel data. Nothing in the workflow above touches a competitor's systems or private accounts. The place to be careful is bidding on a rival's trademark in ads, which is governed by platform policy and trademark law rather than by the research.

How accurate are competitor keyword estimates?

Accurate enough to rank opportunities and choose between them, which is what you need. Positions are closest to reality, volumes are averaged and banded, and traffic figures are modelled. Use the first to decide, the second to size, and the third only to sort.

Can I see the keywords a competitor is bidding on?

You can see the ads they are running in the Ads Transparency Center, and the advertisers overlapping with you in auction insights. Keyword level paid data from a third party tool is an inference from observed ads, so treat the themes as reliable and the individual terms as a lead to verify.

Do I need a paid tool for this at all?

Not to start. Auction insights, Keyword Planner and your own Search Console will fill a first plan, and they cost nothing. Pay for a platform when you need competitor level ranking exports regularly, or when several people need the same numbers.

How often should this be repeated?

Quarterly for the full study, monthly for a short check on the three competitors and the clusters you are actively working. Keep each export, because the comparison between two quarters is more informative than either quarter alone.

The queries that will not appear in any export

Every method above assumes a buyer typed a term into a search box and received a page of links. A good deal of the same research is now a conversation instead. There is no keyword in a conversation and no results page to crawl, so when an assistant tells somebody which supplier to call, that recommendation leaves no trace in any keyword database. A rival can be the standing answer to your best buying question and remain invisible in every export you run, however large the tool's index.

The research is still perfectly doable; the unit changes from a term to a question. Write the ten questions somebody would ask before choosing in your category, leave your own brand name out of the wording so the answer is not steered, put each one to the assistants, and keep a record of the companies named and the pages they were picked up from. AI Knows Us was built to do exactly that, question by question, and it shows where each recommendation was picked up. The output is a worklist in the same shape as the clusters above, which is the useful part: you already know what to do with a list like that.