What Is ChatGPT Search?

ChatGPT search is the feature that lets ChatGPT look things up on the web and answer with links to the sources it used. When a question needs current information, OpenAI says ChatGPT can turn the request into one or more search queries, retrieve results and write the answer from them. For a health practice, this is the route by which your website can be named or linked in a ChatGPT answer.

How ChatGPT search works

Without search, ChatGPT answers from what it learned in training, which stops at a past date and can be out of date. That fixed snapshot is its training data. With search on, it can read current pages. A patient who asks, "Which hormone clinics near me offer bioidentical therapy?" can get an answer assembled from live results, with a Sources button that opens the list of pages it used. OpenAI says search results and citations can be incomplete, outdated or incorrect, so every answer is a snapshot.

The mechanics rest on three OpenAI crawlers, which are automated programs that read websites. OpenAI publishes what each one does.

• OAI-SearchBot is used to surface websites in search results in ChatGPT's search features. OpenAI says a site that opts out will not appear in ChatGPT search answers, though it may still show as a navigational link. See OAI-SearchBot.
• GPTBot crawls content that may be used in training OpenAI's generative AI foundation models. It is a separate setting from search.
• ChatGPT-User visits a page when a user's question in ChatGPT or a custom GPT prompts it. OpenAI notes that because a user initiates these actions, robots.txt rules may not apply to them.

OpenAI also says these settings are independent, and that it can take about 24 hours from a site's robots.txt update for its systems to adjust. That detail matters if you fix a block today and test tomorrow.

Two background facts explain why this is a live issue. First, bot blocking has become common. Cloudflare announced in July 2025 that it was changing its default to block AI crawlers, which means a site behind Cloudflare may be blocking bots without the owner choosing to. If you or a developer set a blanket rule, you may have excluded the one bot that matters here. The topic is covered under AI bot blocking.

Second, being allowed to appear is only the entry ticket. ChatGPT still has to choose your page. This is the idea behind retrieval-augmented generation, where a model retrieves documents and writes from them. Two sets of data show what it tends to retrieve.

SE Ranking analyzed 129,000 domains and 216,524 pages and found referring domains were the strongest predictor of ChatGPT citations. Domains with up to 2,500 referring domains averaged 1.6 to 1.8 citations, while those with over 350,000 averaged 8.4. That is correlation, not cause. It does say that how widely other sites reference you matters.

MGMA's June 2026 analysis tracked 33,340 ChatGPT citation events across 200 practices in 20 metros and 10 specialties, using GPT-5.5 through the API in May 2026. Its author founded a commercial AI visibility service, so read it as one vendor's data. Cited sources were hospital staff rosters at 27.5%, practice-owned websites at about 23%, specialty association credentialing pages at 22.1% and national medical directories at 8.3%. Review aggregators were 3.8%, and the author reported zero direct citations of Google Business Profile.

An answer that cites your page is an AI citation. Getting there takes a reachable site, a page that answers the question plainly, and corroboration elsewhere.

In short, ChatGPT search has two gates. The first is access, which is a technical setting you control in a text file. The second is selection, which depends on your pages and on what the rest of the web says about you. Practices usually fix the first in an afternoon and spend months on the second.

Example: ChatGPT search at a health practice

This example is illustrative, not a transcript. A hormone clinic in Austin has a clear page on bioidentical hormone therapy. A patient asks ChatGPT, with search on, "Which hormone clinics in Austin offer bioidentical therapy, and what does a first visit involve?"

ChatGPT turns this into a few queries, retrieves results and writes an answer that names three clinics. Each has a numbered link. The Austin clinic is not among them.

The owner checks two things. First, robots.txt. The file contains a line that disallows several AI bots, and OAI-SearchBot is on the list because a plugin added a block of AI crawler rules during a security update. Nobody on the team knew. So ChatGPT's search system was told to stay away.

The owner removes the OAI-SearchBot block, leaves the GPTBot rule in place because the clinic prefers its pages stay out of training, and waits a day, since OpenAI says changes can take about 24 hours.

Second, the owner opens the sources ChatGPT cited and sees the clinics it named appear on a state directory page and a specialty association member list. The Austin clinic's providers are not on either. The action list grows. Claim those profiles, make the clinic name and phone match, and add a plain description of what happens at a first visit to the hormone page.

The owner retests the same question next week and the week after, in a fresh chat each time, and logs the cited pages as well as the names. A single run is a snapshot, so only a source that repeats counts as a pattern. The answer may or may not change, and no one can promise it will. The log shows what moved. This is what ChatGPT search work looks like in practice, a small number of fixable causes found by reading the sources the tool itself shows.

How to check yours in 60 seconds

1. Open ChatGPT, turn on search if you see the option, and ask for the best practices for your service in your city.
2. Click the Sources button and read the cited pages. Note which sites appear.
3. Open yourwebsite.com/robots.txt in a browser and look for "OAI-SearchBot" or a blanket "Disallow: /".
4. Repeat the question once with different wording, and once more in a fresh chat on another day. Wording and sources can change between runs, so log the cited pages each time and look for repeats before drawing a conclusion. The reasons are covered under AI answer variability.

What to do about it

1. Make sure robots.txt does not block OAI-SearchBot, and check any security plugin or CDN setting that adds bot rules.
2. Decide on GPTBot separately, as a training choice.
3. Keep your pages fast, readable and public, with a direct answer in the first two sentences of each service page.
4. Keep your practice details identical on your own site and on the directories and association pages that list your providers.
5. Add a ChatGPT referral filter in your analytics, as covered under ChatGPT referral traffic.
6. Retest five patient questions every month and keep a log of which sources are cited.