AI search
How to track competitor rankings in AI search
Last reviewed 24 August 2026 · 6 steps
The short answer
Run one shared prompt set across the assistants your buyers use, and for every answer record each brand named, each URL cited and which source page earned the citation. Comparing citation share across brands over repeated runs gives you the competitive picture, and the cited pages tell you exactly which format is winning the category.
The method, step by step
Use one prompt set for every brand
Comparability comes from asking identical questions. Write prompts around category, alternatives and evaluation intent, then score every brand against the same runs. Separate prompt sets per competitor produce numbers you cannot put on the same chart.
Cover the assistants your buyers actually use
ChatGPT, Gemini and Perplexity cite differently, and Perplexity in particular leans on recent pages. Running the same prompts across all three shows whether a rival is broadly strong or strong in exactly one place.
Score mentions and citations per brand
Count how often each brand is named, and separately how often each brand's own domain is cited. A competitor mentioned everywhere but never cited is being described by third parties rather than quoted, which is a much easier position to attack.
Identify which page types earn the citations
List the cited URLs by type. In most software categories the recurring winners are multi-vendor roundups, pricing pages and documentation — not homepages. That distribution is your content brief.
Find prompts where the whole category is weak
The fastest wins are prompts where answers cite forums, out-of-date posts or nothing authoritative. A single page that answers that prompt cleanly, with a checkable table, can become the default citation.
Report share over time, with the prompt set attached
Publish citation share per brand per run, and keep the prompt set alongside it. Without the prompt set the numbers are not reproducible, and non-reproducible AI-visibility numbers do not survive scrutiny in a board meeting.
Honest limits
What this method will not tell you
- Assistant answers are non-deterministic, so competitive gaps under a few percentage points across a small prompt set are not real findings.
- None of the assistants publish query volume, so citation share cannot be weighted by demand the way SERP share of voice can.
- A competitor cited by a third-party roundup is not the same as that competitor ranking. Attribute citations to the page that earned them, not to the brand mentioned in the text.
Questions
Frequently asked
How many prompts do I need for reliable competitive data?
Enough that one variable answer cannot move the result — typically 20–50 prompts run repeatedly. Consistency of wording matters more than volume.
Can I see a competitor's AI search traffic?
No. Assistants publish neither impressions nor clicks, and referral traffic is only visible in your own analytics. Citation share is a visibility proxy, not a traffic estimate.
Why do competitors get cited when they rank below me in Google?
Assistants tend to quote pages that state checkable facts plainly, so a competitor with a clear comparison table can be cited ahead of a higher-ranking page that buries its answer in marketing copy.
Keep reading