Competitive analysis compares how brands appear under the same test conditions. Record frequency, accuracy, requirements, and citations. The purpose is to identify a useful action supported by evidence.Research sources and limitations. Identifies the available report data and the analysis still needed to support earlier draft figures.
Use the method from Measuring AI Search Visibility and retain the response records described in Building a Visibility Tracking System.
Benchmark before diagnosing
Count competitor mentions and recommendations across repeated runs of the same prompts. Use consistent scoring rules. Report sample sizes and separate results by product and search condition.Don't Measure Once: Measuring Visibility in AI Search, which shows one-off observations are unreliable and repeated measurement is required to characterize a brand's position.
Compare results by model and customer requirement before combining them. A competitor’s broad average may conceal areas where its product does not fit the customer’s needs.
Compare the test conditions
Compare answers with and without search where supported. The difference identifies an association with retrieval access. It does not reveal the cost or method required to change a competitor’s position.
Inspect the sources in searched answers and the facts supplied about each brand. Check whether a competitor has clearer requirements, more complete specifications, or relevant independent evaluations. Treat each explanation as a hypothesis to verify.
Citation-gap analysis
Collect citations from answers that recommend competing products. Separate brand-owned sources from independent sources. Review the cited passages, not just domain counts.Research sources and limitations. Identifies the available report data and the analysis still needed to support earlier draft figures.
An absent mention on a relevant cited page may be worth investigating. First check the page’s scope, criteria, accuracy, and editorial policy. Citation does not guarantee that inclusion would produce a recommendation.
Where retrieval logs are available, review uncited sources separately. Where they are unavailable, mark that information as unknown. Community discussions can still be assessed for real customer problems.
Find useful work
Look for unmet information needs: unanswered requirements, inaccurate descriptions, and useful comparisons that do not exist. Variation across runs is a measurement finding, not proof that a recommendation is easy to obtain.
Prioritize findings by customer relevance, evidence strength, likely business value, and the work required.
Review changes over time
Apply the same checks to your own business and competitors. Record changes in source pages and known product configurations. Avoid interpreting an isolated shift as a competitor’s successful campaign.
Repeat the comparison on a useful schedule and after significant changes. Keep the prior answers so you can inspect what actually changed. Running the Program covers ownership and review.
Compare competitors within the same answers
Choose the competitors customers actually consider. In each completed answer, score all tracked brands using the same rules. Record newly appearing brands separately until their inclusion in future tracking is decided.
If Company B appears more often for weekend service, inspect what the answers say about availability. Check the cited pages. The useful finding may be that Company B states booking conditions clearly while your page omits them.
Distinguish a confirmed difference from an explanation you have inferred. “Company B’s page states weekend hours; ours does not” is directly checkable. “Those hours caused Company B’s recommendation” requires additional evidence.
Present the comparison as a short action table: customer requirement, observed answer difference, source evidence, proposed correction, and verification method. Avoid a competitor score that conceals all five.