Short answer: Noindex is an intentional discoverability decision and only works when the crawler can read it. For noindex and ChatGPT Discovery, measurement should begin with a data contract, not a dashboard. ChatGPT Search discovery starts with public web accessibility and publisher controls. OpenAI says publishers that want pages included in ChatGPT Search summaries and snippets should allow OAI-SearchBot; blocking it can prevent inclusion in those surfaces.

Define the metric before collecting it

Write five fields for every metric: name, numerator, denominator, observation window and known blind spots. This prevents a citation count, prompt sample or referral session from being presented as if it measured total demand.

Baseline design

Choose a stable group of pages or topics before the intervention. Record discoverability, cited or linked pages, identifiable referrals and conversions. Preserve the same cohort during the first comparison window unless the purpose of the experiment is specifically to change the cohort.

Event taxonomy

Separate events into four layers:

Availability. Crawl, index or source eligibility.

Visibility. Mention, citation, supporting-link or measured appearance.

Engagement. Visit, scroll, return, download or another audience behavior.

Outcome. Lead, sale, subscription, pipeline, revenue or another business target.

The layers can influence one another, but they are not synonyms.

Measurement table for noindex and ChatGPT Discovery

Question Example signal Reporting rule
Are we available? crawl/index state binary or coverage, not a rank
Are we being used? discoverability, cited or linked pages, identifiable referrals and conversions report platform and sample scope
Do users engage? qualified sessions / actions separate known referrals from inferred influence
Does it matter commercially? target conversion use assisted views when last-click is incomplete

Experiment design

Change one meaningful element: source structure, technical access, evidence depth, page ownership or update policy. Record the date. Give the system enough time to recrawl or re-evaluate. Compare against the baseline and against a reasonable control group when available.

Do not retroactively choose the metric that moved most. The success criterion belongs in the plan before the result.

Sampling and uncertainty

Many AI visibility tools work from prompt panels or sampled platform data. Report the engine, locale, model or interface when known, prompt set, date range and sample size. “Share of voice” without a denominator is an attractive label, not a reproducible metric.

Executive reporting

A useful executive page contains fewer metrics, not more: availability health, visibility trend, qualified audience behavior and business outcome. Add a short evidence note explaining what is directly observed and what remains inferred.

Conclusion

noindex and ChatGPT Discovery should improve decision quality, not produce more charts. Define the metric contract, protect the baseline, separate visibility from value and report uncertainty explicitly. That makes the data useful even when AI platforms expose incomplete measurement.

OpenAI-specific operating context

ChatGPT Search introduces a crawler-policy question that is easy to confuse with ordinary indexing. OpenAI documents OAI-SearchBot as the crawler used for search discovery and provides publisher guidance around discoverability and citation. That makes crawler access a deliberate publishing decision rather than an invisible default.

The practical implication is to separate three questions: can OAI-SearchBot fetch the page, does the page itself permit the intended form of discovery, and is the content useful enough to be surfaced or cited? These are independent gates. A publisher can make a technically accessible page that is still weak evidence, or publish excellent evidence that a crawler cannot fetch.

Measurement is also bounded. Identifiable referral traffic captures visits, while citations or mentions can occur without a click. A defensible report therefore keeps access, citation/mention observations, referrals and conversions in separate columns rather than turning them into one synthetic “ChatGPT score”.

Applied question for this article

The specific decision is noindex and ChatGPT Discovery. Use the principle in the short answer as the hypothesis to test; document one concrete page, source or workflow where it applies; then record one counterexample or condition where it does not. This keeps the article tied to its own intent instead of drifting into generic AI-search advice.

Sources reviewed

Related insights