how to track a client's brand visibility in chatgpt and perplexity

Start with the decision the client needs to make. A company entering a new country may care about local buying questions; a product team may need to correct inaccurate integration descriptions. These call for different prompt samples. “Improve our AI score” is too vague to define a useful report.

Write down the client’s primary brand, product names, domain, approved aliases, target countries, and language. List the competitors that belong in the comparison. Confirm how you will handle a parent company, acquisition, or name shared with an unrelated business. Keep an ambiguity column instead of forcing every match into yes or no.

Use public product information in checks unless the client has approved another data source and destination. Do not paste confidential customer records, unreleased financial information, or private strategy into consumer prompts merely to make the answer more detailed.

1. Discovery questions: Ask which products solve a specific buyer problem without naming the client.
2. Comparison questions: Ask about a relevant choice between alternatives, with the requirements that determine fit.
3. Capability questions: Test whether the platform accurately describes a named product’s integrations, limits, and supported users.
4. Validation questions: Check important factual claims such as supported regions or implementation requirements.
5. Branded questions: Keep explicit brand queries in a separate group so they do not inflate discovery coverage.

Assign an ID and intent category to every question. Freeze the wording for the reporting period. If the client changes positioning or adds a market, create a new cohort and preserve the previous one for a fair comparison.

Use fresh conversations and record the date, language, visible model or mode, and relevant account context. For ChatGPT, note whether web search was used and whether sources appeared. For Perplexity, note the selected search mode. Save the complete answer and the source URLs, not just a favorable sentence.

OpenAI’s crawler documentation distinguishes OAI-SearchBot, GPTBot, and ChatGPT-User. Perplexity’s crawler guide describes its own search bot and user-triggered fetcher. These are platform-specific controls; checking one crawler does not establish access for the other platform.

If a monitoring service uses APIs, record that fact in the report. An OpenAI API response is not automatically a record of the ChatGPT consumer interface. A Perplexity API response is evidence from that API. Keep the surface label visible in exports and charts.

Put counts beside percentages. In a hypothetical report, “6 of 18 completed unbranded answers mention the client” is easier to assess than “visibility 33.” Include a short note on uncertain brand matches, changed prompts, and changes in the collection method.

A useful monthly summary contains the scope, comparable results, a few inspectable examples, and one or two actions supported by the evidence. Do not fill the report with generic recommendations that could apply to any business.

A spreadsheet suits a small library and lets you refine the rules before committing to software. Its cost is collection and review time. A monitoring service can simplify repeated capture, but the important questions are coverage, surface fidelity, evidence export, and access permissions.

Before selecting a service, ask it to run a small agreed prompt set and show the raw answer behind each metric. Check whether client workspaces are isolated, whether historical exports remain accessible, and how pricing changes with prompts, models, and locations. Our comparison of four tools provides an evaluation method without a paid endorsement.

This site offers a free llms.txt draft generator. Live Perplexity checks are currently unavailable. It does not provide subscription monitoring or claim to measure ChatGPT using Perplexity.

Inspect the cited sources behind an inaccurate or missing answer. If the client’s own page omits a capability or leaves an important restriction unclear, prepare a factual page update with a source the client can verify. Record what changed and when it was published.

Recheck the same questions afterward and show the result even when it is unchanged or inconclusive. Changes in AI answers can have several causes, so avoid assigning every movement to your own work. Measure actual visits through GA4 source reporting, then assess relevant key events. A visibility report should not claim conversions that were never observed.