How do I block ChatGPT without hurting my Google traffic?

A company name is the wrong unit for a rule

The question usually starts with company names: keep Google, block OpenAI. That is understandable, but it does not map neatly to the requests your server receives.

One operator may index your site for search, fetch a page for a live answer and collect content for training. You may value each activity differently. Blocking the whole company treats all 3 as worth zero.

Mixed fleets are most of the problem

In July 2026, Cloudflare reported that operators using one fleet for search, agent traffic and training collection account for over 36% of automated activity. That is more than a third of the traffic these rules need to address.

Read each major operator's crawler documentation before writing rules. Check it again in 6 months. The crawlers and their purposes change, so a list from last year may no longer apply.

Write the rule against the activity

Specify the operator and the activity together. You might block one operator's training collection while allowing its search indexing. Agent retrieval may be worth allowing because a reader requested it. Make a separate decision about grounding, which sometimes produces a citation.

This requires traffic classified by both operator and activity. Supertab Connect does that at your existing CDN edge. With both identified, you can write a precise rule.

Where the rule gets applied

Apply the rule at your own edge. Supertab Connect identifies and reports traffic; it does not block it. The decision to refuse a named company belongs to you, and you should retain control of it.

In practice, configure a rule in Cloudflare, Fastly, CloudFront or your existing provider using the classification results. On WordPress, the free Supertab plugin provides reporting without changes to your CDN configuration.

Check afterwards that it did what you meant

After applying the rule, watch the same report for 2 weeks. Confirm that the activity you wanted to stop has stopped and that the traffic you wanted to keep still arrives. This check is often skipped.

Roughly half the time, something unexpected appears. A crawler thought to be collecting training data was feeding search, or an unfamiliar operator was more active than the one you blocked. The report lets you find this before it shows up in a traffic graph 3 months later.