AI Search

How to compare your brand and competitors in AI answers

Branddi · Published on August 26, 2026 · Updated on October 22, 2026

To compare your brand with competitors in AI answers, use a fixed set of confirmed companies and classify each answer by mention, recommendation, source and appearance order. Calculate indicators from valid, comparable observations. Preserve the passages supporting each interpretation, without treating an AI-generated list as an official market ranking.

A company mentioned first may simply introduce a category. A publisher can appear as a source without selling the solution. Another brand may receive a recommendation without a link to its website. These differences change both the interpretation and the editorial action.

This guide focuses on company comparison and classification. To build and run a questionnaire, use the question set for monitoring your brand in ChatGPT. Avoid changing the test whenever you encounter an unfavorable result.

Which companies belong in the comparison?

Define the category, audience, market and need being addressed. Then select competitors that serve that need. A consultancy, a software tool and an advertising platform can appear in one answer without offering equivalent services.

Maintain an entity register with official names, domains, alternative names, relationships to your business and supporting evidence. Record who reviewed each entry and when. A domain helps resolve ambiguity, but an unfamiliar URL still needs verification.

• Observed entity — Possible classification — Required verification — Include in the competitive set?
• Provider in the same category — Confirmed competitor — Compatible offering and market — Yes, after review
• Publisher of a provider list — Editorial source — Page identity and role — Not automatically
• Commercial partner — Partner — Current relationship and scope — Only with an explicit rule
• Similar company name — Namesake — Domain and context — Not until clarified
• New name in an answer — Entity under review — Identity and activity — Exclude until reviewed

Do not convert every detected entity into a competitor. Keep newly discovered names in a separate research list outside the set used for rates. This allows market discovery without silently changing the historical comparison base.

How do mention, recommendation, source and order differ?

A mention records the correct company name in the answer. A recommendation requires a passage presenting it as an option for the stated need. An owned source requires a displayed URL belonging to a confirmed official channel. Order describes where the entity appears under a documented rule.

One answer can meet all four criteria, some or none. Do not hide these combinations inside a single score. Count each company once per answer for each indicator, even when its name repeats or several links lead to the same domain.

ChatGPT documentation explains that search results and citations appear when web search is used: https://learn.chatgpt.com/docs/web-search, accessed October 22, 2026. Record the link actually displayed; similar content elsewhere does not prove that another page was used.

When examining sources, distinguish a company's own channel from a third-party page about it. A news article may support a recommendation without being an owned source. Check that the link refers to the identified entity and topic.

Does appearing first mean being the preferred option?

Not necessarily. Define whether order means the first textual mention, position in an options list or position within a specific answer section. Do not mix a name's position in the text with a link's position in the source panel.

A list may be alphabetical, group specialties or simply enumerate examples. Look for selection language and reasoning. Without explicit preference, record “observed order, no demonstrated preference.” Missing evidence does not need to be replaced by an assumption.

Google explains that AI Mode and AI Overviews can use different models and techniques, producing different answers and links: https://developers.google.com/search/docs/appearance/ai-features, accessed October 22, 2026. Compare each product separately before combining results.

What denominator should each indicator use?

Use a consistent unit throughout the period. For an answer-based rate, the denominator is the number of valid answers in the selected scope. Collection failure or incomplete content does not mean brand absence. Show expected, valid and unresolved observations.

These formulas are a transparent operational proposal, not an official platform score:

• Company presence: valid answers mentioning the entity divided by valid answers.
• Company recommendation: valid answers with a demonstrated recommendation divided by valid answers.
• Owned source: valid answers displaying at least one official link divided by valid answers.
• Competitor-only presence: valid answers mentioning that competitor but not your brand divided by valid answers.

For a share across brands, fix the company set and count distinct answer-company pairs. The denominator becomes the total number of those pairs across all included companies. That share differs from answer-based presence. If no company in the set appeared, report the share as not calculable rather than inventing a division by zero.

Always name the measured signal. A share based on mentions must not be labeled as a share of recommendations. Readers should be able to reproduce the calculation from the underlying observations and the declared company set.

How can an example make the calculations clear?

Consider a wholly fictional example unrelated to Branddi or customer results. A round schedules twelve observations: ten are valid and two fail. The competitive set contains only Brand A and Brand B, both previously confirmed.

• Indicator across ten valid answers — Brand A — Brand B
• Presence — 4/10 = 40% — 6/10 = 60%
• Demonstrated recommendation — 2/10 = 20% — 3/10 = 30%
• Displayed owned source — 1/10 = 10% — 2/10 = 20%
• Answer-company pairs with presence — 4 — 6

Suppose both brands appear together in three answers. Their union covers seven answers; three valid answers mention neither. Presence share across the two companies is 4/(4+6), or 40%, for A and 60% for B. Its numerical match with the presence rates above is merely a consequence of the chosen example.

If a third company joins the set, share can change without any change to previous answers. Preserve the earlier version and recalculate both comparison periods using the same set when sufficient data exists. Otherwise, identify a break in the series.

How can the comparison remain fair over time?

Compare the same question, intent, platform, mode, language and market. Separate questions that omit company names from those that provide them. An answer prompted by “A or B?” does not measure spontaneous discovery.

Preserve complete answers, dates and known settings. Do not claim that unknown conditions are identical. If one period contains more observations from one product, aggregate results may shift because of composition. Show separate segments first and explain any weighting.

AI Mode guidance warns that answers can contain errors: https://support.google.com/websearch/answer/16011537?hl=pt-BR, accessed October 22, 2026. Verify a capability attributed to a competitor before turning it into a commercial conclusion.

How should ambiguous classifications be reviewed?

Keep the exact passage and classification rationale. If identity or recommendation is unclear, mark that field for review. Do not convert uncertainty into either absence or confirmed presence.

Use human review for cases that affect the report's conclusion. When correcting a classification, retain the previous value, new value, reason and reviewer. Recalculate affected indicators and version the report so that a rule change does not look like a change in AI behavior.

Record elements carrying a commercial label separately. Classification should follow the observed evidence; do not assume sponsorship simply because a company is prominent. Likewise, do not count an identified advertisement as an editorial recommendation in the answer.

How can comparison guide a content decision?

Start with the demonstrated gap. If a competitor is associated with a real capability that your company also offers, check whether the corresponding page explains it accurately. If the difference is a real product limitation, changing the wording should not conceal it.

Connect question, company, passage, source, hypothesis and existing page. Update the page already serving that intent before creating a competing URL. Define what the next review will examine: a more accurate description, an owned source or a relevant recommendation.

The AI Visibility solution evaluation guide explains monitoring requirements. To separate presence from revenue, consult AI Visibility ROI. Several pages changing close together limit attribution to any single publication.

Which questions help interpret the comparison?

Should every detected name enter the ranking?

No. Confirm the entity and its relationship to the category. Publishers, partners and namesakes need separate treatment.

Can mention and source be added as two results?

Keep them as separate indicators. One answer can contain both; adding them without an explicit formula combines overlapping signals from the same observation.

Does one round establish a trend?

It describes the observed sample. A trend requires a comparable series with visible failures and method changes. No universal collection frequency fits every objective.

How can we evaluate this monitoring with Branddi?

Bring your company list and priority questions. Branddi AI Visibility, accessed October 22, 2026, describes monitoring of mentions, sources and competitors within an agreed scope. Request a demonstration connecting each indicator to its supporting answer and classification.

Back to blog