Most competitor analyses end as a slide deck: a table of domain ratings, a keyword overlap chart, a list of things the competitor does. The deck is filed, and nothing changes. A useful analysis ends somewhere else: with a short, ranked list of pages to create, fix or get onto, each tied to a question your buyers ask.
This post gives the method for Google and for AI answers, the sheet we use, and the numbers from our own head-to-head against the brands that win our category.
Who are your SEO competitors?
Not necessarily the companies you lose deals to. Your SEO competitors are whoever wins the searches and answers your buyers see. For most queries that list contains three kinds of site:
Business rivals
- Example
- Companies selling what you sell
- What to do about them
- Out-answer them on your own pages
Publishers and roundups
- Example
- Review sites, "best X" lists, trade press
- What to do about them
- Get onto their pages
Reference sources
- Example
- Docs, encyclopedias, platform help centres
- What to do about them
- Rarely beatable; cite them, complement them
Separating them early saves wasted work. You cannot outrank a platform's own documentation for a question about that platform, and you do not need to. You can get onto the roundup that outranks your rival.
How do you run the analysis?
Step 1: Write the question list
Thirty to sixty searches and questions your buyers ask, across the buying stages: problems ("how do I…"), categories ("best X for Y"), comparisons ("X vs Y", "X alternatives") and brand questions. Use buyers' words. This list is the backbone of everything that follows.
Step 2: Record who wins on Google
For each search, record the top ten results, which of them are business rivals, which are publishers, whether there is a featured snippet or AI Overview, and where you rank. Search Console gives you your own positions; for the rest, search and note it.
Step 3: Record who wins in AI answers
Ask each question on the AI engines your buyers use, several times each. For every answer record which brands are named, in what order, and which pages are cited. Count per brand: named in how many answers, out of how many read.
Step 4: Study the winning pages
For the questions you lose, open the winning pages and ask why they win:
- Do they answer the question directly, near the top?
- Do they have something you do not: original data, a comparison table, a clearer definition, a tool?
- Are they newer, or more complete?
- Who links to them, and who cites them in AI answers?
Step 5: Turn it into work
For each lost question, one of four moves:
- Fix our page: you have a page; it answers worse.
- Publish our answer: you have no page for it.
- Get onto their page: a roundup or guide decides it, and you are not on it.
- Leave it: a reference source owns it, or the question does not matter enough.
Rank the moves by the value of the question and the cost of the move. That ranked list is the deliverable.
What does a head-to-head look like on real numbers?
Here is ours. Our workspace tracks 61 questions about our category and reads them daily on ChatGPT and Perplexity. The head-to-head below is a read taken on 2026-10-04 at about 18:20 UTC: from 2026-09-04 to 2026-10-04 (23 of 31 days sampled) it covered 4,374 answers. For each competitor, it counts the answers where only we were named, only they were named, both, or neither.
A later read the same evening covered 24 of 31 days and 4,524 answers. It gives brand totals but no head-to-head split: Semrush named in 1,621, Profound 1,420, Ahrefs 1,171, Peec AI 988, Otterly AI 967 and Attensira 197 (the totals how to calculate share of voice works through). The table keeps the earlier read so its four columns come from one read.
The table at the top shows four of the five counts per competitor, all from the 18:20 UTC read. The fifth, answers that named neither of us, ran from 2,627 for Semrush to 3,268 for Otterly AI. Three things the counts say that a keyword overlap chart would not:
- We lose far more than we win on category questions. Every competitor here was named alone in hundreds more answers than we were.
- Profound is the competitor we are most often named beside: 78 answers named both. Those are mostly comparison questions that contain both names, which tells us where buyers already put us side by side.
- The "neither" count is the opportunity. For every competitor, thousands of answers named neither of us. Those answers were won by someone else, often a publisher or a smaller tool, and they are not locked up by a big rival.
A head-to-head like this is a rate count, with the same total behind every row, so it does not shift when you add or remove a competitor. See how to calculate share of voice for why that matters.
Which pages decide the answers your competitors win?
The most useful part of an AI competitor analysis is not who wins but which pages make them win. In our own data, the single most cited pages for category questions were vendor roundups and comparison hubs. That told us the move: not out-ranking Semrush's domain, but appearing on the roundups and publishing comparison pages worth citing.
To find these pages, collect every URL cited in the answers your competitors win, count how often each appears, and sort. The top of that list is your outreach list and your content brief at the same time.