Build a fair competitor comparison.

Start with the buyer’s shortlist.

Choose brands a buyer could reasonably consider for the same job. A direct alternative with a similar audience is usually easier to interpret than a large company that happens to share a category keyword. Include a broader alternative when buyers genuinely compare it, and record why it belongs in the set.

Begin with a handful of competitors you can review carefully. The aim is a useful comparison, rather than filling every available tracking slot. Keep a separate research note for names you are still evaluating.

• What problem does the buyer want to solve?
• Which audience, market, or product category does this comparison cover?
• Would a buyer consider these brands in the same decision?
• Are you comparing a whole company or a specific product?

Define the names before counting mentions.

Use the brand name and its actual website domain. Add genuine alternate names as aliases when the product has a shortened name, an established abbreviation, or an older name buyers still use. Avoid generic category words that could match unrelated products.

For a fictional product called Northstar Metrics, an established product abbreviation could be a useful alias. A broad word such as analytics would not identify that product. Read the supporting quote to confirm that the answer refers to the company you intended.

Keep parent companies and individual products distinct when they represent different buying choices. A parent-company mention does not necessarily recommend the specific product you track. Review ambiguous evidence before treating it as a competitive win.

Give the brands the same opportunity to appear.

Use a consistent question set and provider mix when comparing brands. Category questions reveal discovery, evaluation questions reveal trade-offs, and branded questions reveal how a provider describes an existing option. These question types answer different research questions.

A question that explicitly names one competitor gives it a different opportunity to appear than an unbranded category question. Keep those contexts visible when reading aggregate results. Do not interpret a higher mention rate on branded prompts as evidence of stronger unprompted discovery.

Compare equivalent periods and review collection coverage. If ChatGPT has completed a question while Claude and Gemini are still waiting, you have an incomplete round. Failed or missing collections are not confirmed brand absences.

Compare assessments from the same answers.

Check the assessed-answer denominator for every brand. Suppose your brand appears in eight of 10 assessed answers, while a competitor appears in 12 of 20. The rates are 80% and 60%, but those totals alone do not establish how the brands performed in a shared set of responses.

Where a comparison uses answers assessed for both brands, inspect the four outcomes: your brand only, competitor only, both mentioned, and neither mentioned. These outcomes separate direct differences from conversations where both brands participate.

A mention is also different from a recommendation position. A brand can appear in a caveat or an unordered list with no explicit rank. Review position, sentiment, and evidence separately rather than converting every appearance into an endorsement.

Treat changes to the set as changes to the measurement.

Record when you add or remove a competitor, change its aliases, or substantially change the question set. Brand settings apply to future tracking; do not assume a saved historical assessment has silently changed to use a new identity.

After an identity change, inspect the brand details and evidence on saved answers before combining old and new results. A previous snapshot can describe the names that were tracked when it was analyzed. Missing historical assessment is different from a historical answer that explicitly omitted the brand.

Use a saved view for a repeatable provider, date, and question scope. Revisit that scope during your review rather than comparing whichever filters happen to be open on different days.

Turn a difference into a focused investigation.

Pick one question where a relevant competitor appears and your brand does not. Open the full saved answer, check its recommendation order, and inspect any cited sources. Ask what the response emphasizes: category fit, use case, audience, integrations, pricing, or another concrete detail.

Write a small next action tied to that evidence, such as clarifying a product capability on your website or reviewing whether a cited page describes your category accurately. Keep the hypothesis separate from the observation.

Collect a comparable follow-up sample before claiming improvement. AI answers vary, and a change in the response does not by itself prove that your edit caused it. A useful review ends with an evidence-backed question to investigate next.