The short answer
Monitor the AI platforms that your eligible buyers plausibly use for category discovery, comparison, and fact verification. Start with direct evidence from customer research, sales conversations, referral data, and a small cross-provider pilot. Score each provider surface on buyer relevance, decision relevance, market availability, observable context, and the cost of collecting reviewable answers. Keep a stable core, test uncertain providers separately, and watch the rest. A ChatGPT-and-Gemini snapshot is a useful first comparison, not proof that those are the only providers that matter to your market.
Editorial method · evidence boundaries
This guide proposes a review method. Worked examples and pilot sizes are illustrations unless a dated study is explicitly identified. Provider documentation supports only the behavior and product surface it describes; it does not prove that an AIBL workflow improves recommendations or business outcomes.
Our historical collectors used recorded API models. Those observations do not establish what every consumer ChatGPT, Gemini Apps, Claude or Grok user sees. Record the product surface as well as the provider.
Historical AIBL records have limitations: completed jobs can contain truncated answers, and native citation metadata was not retained uniformly. Verify answer usability and source availability before scoring or comparing results. Missing saved metadata is not proof that a provider supplied no sources.
What businesses notice
Common signs of the problem
- The team tracks every available provider but cannot explain which buyer decision any of them represents.
- A platform was selected from a broad popularity claim that does not describe the company’s audience, market, account type, or use case.
- Results from consumer chat, web search, APIs, connected apps, and research modes are pooled as though they were the same product surface.
- Provider coverage keeps changing, so trend lines move even when the brand’s observed answers have not materially changed.
01
Choose provider surfaces, not logos
“Track ChatGPT” is incomplete scope. The same company can offer consumer chat, web search, mobile assistants, enterprise workspaces, APIs, and connected tools. Name the surface and its observable conditions before calling it one measurement channel.
Define the surface a buyer can actually use
Record the product surface, account state, device when material, market, language, search or grounding state when visible, and collection date. A public consumer answer is not interchangeable with an API response, an enterprise workspace, a deep-research workflow, or a personalized assistant.
Treat web access as context, not a provider constant
ChatGPT may search automatically or when a user selects Search. Claude web search can be enabled or disabled and may depend on workspace settings. Gemini may show sources or related links for some responses. Grok can decide whether to search public X posts or the web. Preserve the observed state instead of labeling an entire provider “web-grounded.”
Review ChatGPT Search behavior ↗Do not infer unseen retrieval
A citation, source panel, or search indicator is observable evidence. The absence of one does not prove that no retrieval, product data, location signal, memory, or training knowledge influenced the answer. Report the interface evidence you can inspect and leave hidden causes unknown.
Version meaningful surface changes
If a provider changes the model, search behavior, citation interface, account requirement, or product mode used for collection, create a method version. Do not splice unlike surfaces into one trend merely because the provider name stayed the same.
02
Start with evidence of buyer relevance
There is no universal provider list for every brand. Build the portfolio from observed buyer behavior and business exposure before using outside market summaries as secondary context.
Ask buyers where AI enters the decision
Use customer interviews, win-loss reviews, sales and support notes, user research, and on-site surveys to learn whether people ask assistants to discover the category, compare options, check a claim, plan a purchase, or summarize research. Capture the provider and surface only when the buyer can identify them reliably.
Inspect owned demand signals carefully
Review referral sources, shared conversation links, form responses, sales-call language, and customer-success notes. Referral traffic can confirm some clicked visits, but it cannot count answers that influenced a buyer without producing a click. Keep observed use separate from inferred influence.
Measure AI traffic and leads without overclaiming →Segment before drawing a conclusion
Provider relevance can differ by buyer role, region, language, device, company policy, and decision stage. A tool used by technical evaluators may be immaterial to consumers; a mobile assistant may matter locally while a desktop research mode matters for procurement. Preserve those strata instead of publishing one unsupported audience-wide percentage.
Use outside usage data as a hypothesis
Third-party traffic, survey, app-store, or market-share estimates can suggest providers to test, but their population, period, geography, denominator, and definition of use may not match your buyers. Do not let a broad ranking override direct evidence from the market you are trying to understand.
03
Score each candidate against five selection tests
Use a short provider-selection record rather than a popularity league table. The scorecard is a prioritization aid: publish the inputs and judgment, not just a total.
1. Buyer relevance
State the evidence that eligible buyers use or may use this surface: direct research, a recurring sales signal, referral evidence, documented workflow fit, or an explicitly labeled exploratory hypothesis. “It is a major AI brand” is not enough on its own.
2. Decision relevance
Name the decision the surface could affect and the question families that represent it. Category discovery, comparison, local selection, product verification, and post-purchase support are different jobs. Exclude cells where the tested brand is not legitimately eligible.
3. Market and access fit
Verify that the product mode is available to the intended market, language, device, account type, and workspace policy. Google and Anthropic both document feature variation by settings or availability; X documents Grok as an X product with access tied to where X is available. Recheck these conditions when the portfolio is reviewed.
Review Gemini app availability ↗4–5. Observability and operating cost
Determine whether collectors can preserve the exact question, complete answer, visible sources, timestamp, failure state, and relevant context consistently. Then estimate the question × provider × market × repeat workload, human review time, access requirements, and evidence-retention limits. A relevant surface can stay in a test tier until the method is reliable.
04
Understand what the four major surfaces can change
These current first-party descriptions explain why provider selection matters. They do not predict which provider will mention a brand, prove source quality, or establish audience share.
ChatGPT
OpenAI documents that ChatGPT can search the web automatically or through an explicit Search action, may show citations and a Sources panel, and may use approximate location for local results. Search results and citations can still be incomplete, outdated, or incorrect. A ChatGPT cell should record whether search and location context were observable.
Read OpenAI’s current Search guide ↗Gemini
Google documents that Gemini may show sources and related links, while not every response includes them. Its connected services and capabilities can vary by product, account, device, language, and location. Record the actual Gemini surface and visible source state instead of assuming every run has the same Google context.
Read Gemini source guidance ↗Claude
Anthropic documents web search as a controllable capability that can use live web content and return citations when invoked. Workspace owners may need to enable it, availability can depend on configuration, and Anthropic advises verifying important information. Separate searched and non-searched Claude observations.
Read Claude web-search guidance ↗Grok
X documents that Grok can decide whether to search public X posts and conduct real-time web search. That can make X-native conversation context material for some brands and irrelevant for others. X also warns that Grok can provide incorrect or incomplete information, so preserve and verify the complete answer.
Read X’s Grok documentation ↗05
Build a core, test, and watch portfolio
A tiered portfolio protects trend continuity while leaving room to learn. It also prevents a newly added provider from changing the headline denominator without an explicit methodology break.
Core: decision-relevant and repeatable
Place a surface in the core when buyer evidence is credible, the represented decision matters, market access is stable enough, and collection produces reviewable records. Freeze the provider-surface definition and report core results with a stable denominator.
Test: plausible but not yet proven
Use a bounded pilot for surfaces with plausible buyer relevance but uncertain demand, access, reproducibility, or action value. Keep test-tier observations out of the main trend until the team predeclares the evidence required for promotion.
Watch: review on a trigger
Maintain a lightweight watchlist for providers or modes that are currently ineligible, inaccessible, unsupported, or low priority. Define triggers such as repeated buyer mentions, a new market launch, a material product change, new referral evidence, or a change in company policy.
Retire with a record
A provider can leave the core when buyer relevance disappears, the surface is discontinued, access becomes unreliable, or the result no longer informs a decision. Preserve the historical method and retirement date. Do not rewrite prior reports as if the provider had never been included.
06
Run a fictional 24-answer selection pilot
This example shows how to test provider utility. It contains no measured provider, customer, or AI Brand Lens outcome and cannot establish market share or a universal provider order.
Declare the candidates and questions
Suppose a U.S. software company has evidence that buyers use ChatGPT and Gemini, a recurring but unverified Claude signal, and no direct Grok evidence. Choose three materially different buyer questions: category discovery, role-specific comparison, and a branded fact check. Run the same eligible questions on all four named consumer surfaces.
Make the attempt count explicit
Three questions × four provider surfaces × two fresh-context repeats create 24 planned attempts. Record completed, failed, blocked, excluded, and retried attempts separately. Two repeats help expose obvious instability; they do not estimate every answer a future buyer could receive.
Review evidence and decision value
For each completed attempt, capture brand inclusion, recommendation role, material facts, competitors, visible citations or sources, search state when observable, and whether the answer creates an actionable content, product, sales, or risk decision. Review factual accuracy independently from brand visibility.
Assign tiers without ranking the providers
Suppose the method is reliable for three surfaces and unreliable for one, while buyer evidence is strong for two and only hypothetical for the others. Keep the two evidenced surfaces in core, place the relevant but uncertain surface in test, and place the unsupported or unreproducible surface in watch. The conclusion is about this company’s declared scope—not provider quality.
07
Report provider coverage without hiding the denominator
Provider choice is part of the measurement method. Readers should be able to see what was included, what was not, and how a portfolio change affects interpretation.
Publish the provider-surface ledger
For every tier, show provider, product surface, account state when relevant, market, language, search or grounding state when observable, question families, repeats, collection method, evidence-retention fields, buyer-evidence basis, and last review date.
Show provider results separately first
Report planned attempts, completed answers, failures, mentions, eligible recommendations, citations, and material factual issues by provider surface before any combined number. A pooled score can hide a strong result on one provider and a total omission on another.
Interpret provider disagreement correctly →Treat portfolio changes as method changes
Adding, removing, or replacing a provider changes the sample. Preserve like-for-like core trends and display expanded-portfolio results as a new view. OpenAI evaluation guidance recommends representative test cases, explicit criteria, logging, and continuous evaluation; it does not make unlike provider samples directly comparable.
Review evaluation best practices ↗Start with one bounded comparison
The free AI Visibility Snapshot runs one buyer-style question through ChatGPT and Gemini and preserves both answers for review. Use it to test whether a two-provider disagreement creates a real decision. Expand to Claude, Grok, more questions, markets, or repeats only when the added coverage answers a declared business question.
Run a free AI visibility snapshot →Continue the research
Related AI visibility guides
Primary sources
Primary documentation used for this guide
AI products, search behavior, and platform policies change. Check these maintained first-party sources before making technical decisions.
- Searching the web with ChatGPT ↗
OpenAI Help Center
First-party documentation of ChatGPT Search availability, automatic and manual search, source presentation, approximate location, publisher eligibility, and verification limits. Product behavior can change; the guide uses it only to define observable collection context.
- View related sources from Gemini Apps ↗
Google Gemini Apps Help
First-party documentation that Gemini may show sources or related links and that not every response includes them. Google distinguishes displayed source relationships from its separate response-checking behavior.
- Gemini mobile app availability ↗
Google Gemini Apps Help
First-party availability documentation showing that device, account, language, country, and feature conditions can matter. The guide does not convert availability into evidence of audience usage.
- Enable and use web search ↗
Claude Help Center
First-party documentation of Claude web search, citations, workspace controls, web fetch, usage limits, and verification guidance. It supports separating searched and non-searched collection conditions.
- About Grok ↗
X Help Center
First-party description of Grok on X, including its ability to decide whether to search public X posts and the real-time web, current access context, and the warning that outputs may be incorrect or incomplete.
- Evaluation best practices ↗
OpenAI Platform Documentation
First-party guidance on task-specific objectives, representative cases, explicit criteria, complete logging, continuous evaluation, and human calibration. The core/test/watch provider portfolio is an AI Brand Lens editorial method, not an OpenAI standard.
Measure your visibility
Turn the questions in this guide into an evidence-backed baseline.
Get AI Visibility ScoreTalk to us
Have a visibility question?
Tell us what your team is trying to measure or improve.
Contact AI Brand Lens →Keep learning
Explore every guide.
Browse practical answers about AI visibility, competitors, and measurement.
View all guides →