AI Search Competitor Analysis: An Evidence Review Worksheet

Start with a decision your marketing team can act on, such as which product explanation needs work. Choose the buyer questions and comparison domains before reviewing the answers. Then record what each result actually supports.

A competing domain in a returned source list is a finding to inspect. It does not establish that the competitor won a customer. An unavailable answer tells you nothing about either brand's presence.

This manual worksheet helps in-house marketers, consultants and agencies turn answer evidence they already have into a practical content decision.

Choose one decision and one brand

Write down the decision, the brand's domain and the intended buyer. Choose comparison domains because those businesses are relevant alternatives for that buyer. A publisher or reference site can appear in an answer's sources without being a competing business.

Keep the first review focused. For an agency, complete a separate record for each client brand.

Use a modeled assessment to suggest questions, then check their relevance yourself. A generated competitor gap is a hypothesis to investigate, not evidence that a rival appeared in a real answer.

Fix the question set and collection method

Choose questions that could change a marketing decision. Include the wording a buyer would use for a requirement, constraint or comparison. There is no need to manufacture a long list of near-duplicates.

Before evaluating results, record:

• The exact question wording and planned question count.
• The brand domain and each comparison domain.
• The provider and collection method, including whether it was a consumer-app session, direct API response or third-party collection of web results.
• The location, language, date and available settings.
• The domain-matching rule, including whether subdomains count. Use the same rule for every compared brand.
• The observation population: one result for each chosen question and provider in this review.

Keep different methods separate. A provider API result does not reproduce a buyer's personalized session in the consumer app. If you run a question again, retain the earlier result and identify the new observation instead of selecting whichever answer looks best.

Separate source evidence from other claims

An attributed source is a reference returned with the answer. A wider search-results list may contain candidate pages that the answer did not cite. Use evidence associated with the answer, and keep the provider's attribution label with it.

A source-domain match does not establish a brand mention, endorsement, sentiment, ranking position or website visit. Record those outcomes separately only when you have evidence for them. The KnewSearch methodology explains these distinctions.

Use these states consistently:

• Modeled assessment: generated suggestions or estimates. Useful for planning, but excluded from observed coverage.
• Complete observation, matching source present: the completed answer has usable source evidence containing the tracked domain under the recorded matching rule.
• Complete observation, no matching source: the completed answer has usable source evidence, but it does not contain that domain. This includes an explicitly complete empty source list.
• Pending, failed, unavailable or incomplete evidence: no conclusion about the domain. An empty field alone cannot establish an empty source list.

A missing domain in one returned source list does not show that a site is blocked or that the brand was never mentioned. Investigate Google indexing separately where relevant; Google's AI feature guidance explains its search eligibility requirements without guaranteeing inclusion.

Copy this blank review record

Copy the following fields into your own document. Fill them from actual observations and leave unknowns explicit. No email is needed to use this worksheet.

Review scope

• Marketing decision: [not entered]
• Brand domain: [not entered]
• Relevant comparison domains and why they belong: [not entered]
• Exact buyer questions and planned count per provider: [not entered]
• Provider, method, location, language and settings: [not entered]
• Review date, observation times and domain-matching rule: [not entered]

One observation

Repeat these fields for each question and provider. Repeat the domain-match field for each brand being compared, keeping all brands on the same observation population.

• Exact question: [not entered]
• Provider, method and observation time: [not entered]
• Evidence state: [modeled / complete / pending / failed / unavailable / incomplete]
• Tracked domain and source match: [not entered; yes / no / unknown]
• Attributed source links and evidence reference: [not entered]
• What the evidence supports: [not entered]
• What remains unknown: [not entered]

Review summary and next action

• Planned observations per provider: [not entered]
• Complete comparable observations per provider: [not entered]
• Pending, failed, unavailable or incomplete observations: [not entered]
• Complete observations with a matching source, separately for each tracked domain: [not entered]
• Coverage and completion fraction, or no percentage: [not entered]
• Finding worth investigating: [not entered]
• Proposed page or campaign action and its owner: [not entered]
• Evidence needed to review that action: [not entered]

Show both coverage and completion

Calculate coverage separately for each provider, using one fixed question set and collection method. Count each completed comparable answer once, even if it contains several links from the same tracked domain.

Coverage is the number of completed comparable answers with a matching source for the tracked domain, divided by the number of completed comparable answers, multiplied by 100.

Always display the completed observation count beside the planned count. If some planned observations are missing, label the coverage partial. Do not present it as the complete benchmark or compare it with a fully completed run. If no observations are complete, report no percentage.

Compare each brand against the same completed population and matching rule. Different domains can occur in the same answer, so their coverage percentages need not add to 100. This is answer-source coverage in a defined sample, not market share or the share of all buyer conversations.

Inspect a limited example

The dated KnewSearch sample shows one question observed on September 10, 2026 in United States English through Google AI Overview, ChatGPT web and Gemini web collection. It includes the returned source links and the limits of the finding. Sonar is unavailable in that example.

The sample has no configured competitor comparison. Use it to practice inspecting source evidence, not to rank businesses or infer performance across other questions or markets. It is owned-property evidence, not a customer result.

Turn the finding into one useful action

Inspect a relevant source page or your own page before deciding what to change. You might find an unclear product explanation or an unanswered buyer question. Record the gap you can actually verify, the proposed edit and its owner. A provider's source list alone does not explain why that page was selected.

When you review the action, preserve the earlier evidence and use comparable questions, providers, methods, locations and matching rules. If the question population or comparison set changes, start a new baseline. A changed result is something to investigate; it does not establish that a particular edit caused it.

Keep website visits, qualified leads and paid subscriptions in separate evidence records. None is established by a source match or a favorable coverage number.

Start with a free modeled assessment of your website. Use its suggested questions and gaps to plan your first review, then collect answer evidence for the questions you choose. No email or card is needed.