Direct answer
What should you know first?
- Run the same 20 prompts on ChatGPT, Perplexity, Gemini and Claude in fresh sessions with web search enabled where available.
- Score mention, accuracy, differentiation, citation and actionability separately. A mention with the wrong service description is a failure, not visibility.
- Perplexity is explicitly search-centred and citation-forward. ChatGPT and Gemini have documented web-search or grounding modes. Claude capabilities depend on the product and tools available in the session.
- Test branded and non-branded prompts. Branded prompts measure entity accuracy; non-branded prompts measure whether the brand enters a shortlist.
- Repeat the test monthly because assistants, models and source indexes change.
Platform comparison
How do ChatGPT, Perplexity, Gemini and Claude approach brand discovery?
| Assistant | Documented web mechanism | Strongest testing use | Watch for |
|---|---|---|---|
| ChatGPT | Search can retrieve current web information and show source links | Broad buyer questions and conversational follow-ups | Automatic search choice and query rewriting can vary |
| Perplexity | Search-first answers with citations to original sources | Citation coverage and source diversity | Selected model and search mode can change results |
| Gemini | Google Search grounding can generate queries and return cited responses | Current facts and Google-connected discovery | Grounding supports facts, not guaranteed brand inclusion |
| Claude | Web access depends on the Claude product, plan and enabled tools | Long-context evaluation and structured brand comparison | Do not assume every session has live search |
Decision: Compare the user-visible product and mode you actually care about. A model benchmark does not tell you which sources a live assistant will retrieve for one buyer question.
20-prompt test
Which prompts should you use for an AI brand-discovery audit?
- What are the best providers for [category] in [market]?
- Which [category] provider is best for [specific buyer]?
- Compare five [category] options for [use case].
- Who specialises in [narrow service] for [market]?
- What should I ask before hiring a [provider type]?
- Which provider combines [capability A] and [capability B]?
- What are the risks of choosing a [provider type]?
- Which [category] experts have relevant professional credentials?
- Find a provider for a [size] project with a [deadline].
- Which option is best when direct specialist access matters?
- What does [brand] do?
- Who is [person] at [brand]?
- What are [brand]’s main services?
- What evidence supports [brand]’s experience?
- How is [brand] different from [named alternatives]?
- Is [brand] suitable for [specific project]?
- What do clients say about [brand]?
- Where is [brand] based and which markets does it serve?
- How can I contact [brand] for a quote?
- Which source should I use to verify [brand claim]?
Scoring
How should you score AI brand visibility?
| Dimension | 0 points | 1 point | 2 points |
|---|---|---|---|
| Mention | Brand absent | Appears only after branded prompt | Appears in relevant non-branded shortlist |
| Accuracy | Wrong identity or service | Partly correct with omissions | Correct service, market and audience |
| Differentiation | Generic or confused | One weak differentiator | Specific supported differentiators |
| Citation | No source or wrong source | Relevant secondary source | Relevant owned and independent sources |
| Actionability | No useful next step | Generic website reference | Correct service page, profile or contact route |
Decision: A maximum 10-point result is useful only when the prompt, date, model, mode, location and sources are recorded. Do not turn it into an unsupported market-share claim.
Source readiness
What makes a brand easier for AI assistants to understand and cite?
- One consistent organisation name, person name and service description.
- Dedicated pages for each real service and buyer problem.
- Visible authorship, credentials, dates and contact details.
- Specific claims supported by methods, sources or verifiable records.
- Concise answer passages near the questions buyers ask.
- Consistent profiles on professional bodies and trusted external sites.
- Accessible HTML, crawl permission, stable URLs and internal links.
- Unique case evidence or testimonials with clear attribution.
- Correct structured data that matches visible content.
- A correction process for stale facts, broken links and changed services.
Failure diagnosis
Why might one assistant omit or misdescribe a brand?
| Failure | Likely cause | Evidence to inspect | Fix |
|---|---|---|---|
| Brand absent | Weak category association or inaccessible pages | Non-branded prompt sources and crawl status | Strengthen focused service pages and external corroboration |
| Wrong company | Name collision or inconsistent entity details | Titles, profiles and structured data | Use stable names and sameAs references |
| Outdated service | Stale indexed or cited source | Source dates and cached descriptions | Update canonical pages and trusted profiles |
| No citation | Answer mode or weak source fit | Platform mode and retrieved links | Publish more citable first-party evidence |
| Generic summary | Page copy lacks specific proof | Claims, numbers, credentials and case details | Add verified differentiators with context |
Decision: Fix the source of the confusion, not the assistant answer alone. The answer is a symptom of the information environment it retrieved.
Operating cadence
How should a small brand monitor AI discovery every month?
- Freeze the 20-prompt set and define allowed market variations.
- Run fresh sessions on all four assistants in the same week.
- Save answers, citations, model names, modes and dates.
- Score the five dimensions and flag factual errors immediately.
- Group cited domains into owned, professional, media, directory and low-quality sources.
- Prioritise fixes that affect buyer trust or contact accuracy.
- Update the strongest canonical page and relevant external profile.
- Re-test only after sources have had time to be crawled or retrieved.
- Compare three-month trends rather than one volatile run.
- Connect visibility changes to branded search, referral traffic and enquiries.
Practical verdict
Which AI assistant matters most for brand discovery?
The assistant your buyers actually use matters most. If you do not know, test all four with equal discipline and use referral, sales-call and customer research to adjust the weighting.
Perplexity is useful for citation-heavy diagnostics. ChatGPT and Gemini are important broad discovery surfaces with documented search connections. Claude is important where your buyers use it for long-form research or comparison. The brand should be understandable and verifiable across all of them rather than optimised for one screenshot.
Buyer questions
Frequently asked questions
Can I check AI brand visibility with one prompt?
No. One prompt measures one wording, session and moment. Use at least branded, non-branded, comparison, trust and contact prompts.
Should I include my brand name in the test?
Use both. Branded prompts test entity accuracy, while non-branded prompts test whether the assistant discovers the brand for a real category need.
Why do AI assistants give different brand answers?
They may use different models, tools, search queries, indexes, source-selection methods, personalisation and conversation context.
Are citations the same as recommendations?
No. A page can be cited to support one fact without the brand being recommended. Score mention, sentiment, accuracy and citation separately.
How often should I rerun a brand-discovery audit?
Monthly is reasonable for an active brand or changing category. Use a faster cadence after a major launch, rebrand or material factual correction.
Can structured data force assistants to mention a brand?
No. Correct structured data can clarify entities and page meaning, but it cannot force retrieval, citation or recommendation.
Evidence base
Sources and methodology
This guide uses official product documentation, professional bodies and public authorities where available. Comparisons and scores are the agency's editorial framework, not a claim of official endorsement.
- OpenAI ChatGPT Search Official explanation of web search, query rewriting and source links.
- OpenAI answer reliability guidance Official advice on limitations and verification.
- Perplexity: How does Perplexity work? Official search, synthesis and citation description.
- Perplexity source labels Official explanation of source labels and their limits.
- Gemini grounding with Google Search Official search-grounding workflow and citations.
- Anthropic Claude documentation Official product and platform documentation.
Last reviewed on 4 September 2026. Product features and search behaviour can change, so verify the linked documentation before making a high-risk decision.
Keep exploring
Related specialist guides
About the author