How to Get Your Brand Cited by ChatGPT (Start With Bing)
ChatGPT's live search grounds mostly on Bing's index, so getting cited starts in a tool most marketers have never opened. Eight moves in order, with effort estimates, from the agency that runs them for every client.

TL;DR
- ChatGPT answers in two layers. Retrieval finds candidate pages, mostly through Bing's index; the model then picks which ones to quote. You have to pass both.
- Two robots.txt entries decide whether you're a candidate at all: OAI-SearchBot and ChatGPT-User. GPTBot is a training decision and doesn't affect citations either way.
- Once you're findable, the model favors pages that open with a plain answer. Bylines, publish dates, updated dates, and question-and-answer blocks help.
- The biggest lever sits off your site: being named on the pages ChatGPT already trusts for your category.
ChatGPT can only cite what it can find, and for live answers it mostly finds through Bing. The eight moves that get a brand into ChatGPT's answers, in order, with the effort each one takes.
In July 2026 a site with more than 90,000 ChatGPT citations lost all of them in about two days. It hadn't changed anything. Bing removed it from its index after a spam flag; the ChatGPT citations went with it (write-up). Its Google rankings were near zero the whole time.
That's the mechanism in one story. ChatGPT's live search grounds mostly on Bing's index, through OpenAI's partnership with Microsoft. OpenAI has been building its own crawler, and some tests suggest Google's index shows up for paid users (one analysis), so treat Bing as the floor rather than the ceiling. A page Bing hasn't indexed is a page ChatGPT can't cite.
Two layers
When you ask ChatGPT a question, it fans out into several searches and reads the candidate pages that come back. That's retrieval. Then the model picks the pages that answer plainly and drops the rest. That's evaluation. Being indexed gets you into the room. Quotable pages get named. Most AEO advice only talks about the second layer; most failures happen in the first.
The eight moves, in order
Do them in this order. The later ones do nothing until the earlier ones hold.
1. Let the right bots in. 30 minutes. OpenAI runs three crawlers (docs). OAI-SearchBot builds the index behind ChatGPT search. ChatGPT-User fetches a page live when someone asks about it. GPTBot collects training data. Only the first two affect citations. Your robots.txt needs User-agent: OAI-SearchBot and User-agent: ChatGPT-User, each followed by Allow: /. Then check the edge. Cloudflare changed its defaults on September 15, 2026 to block Agent and Training crawlers on ad-supported pages for new domains and untouched Free accounts (announcement); its July 2025 default had blocked all known AI bots on new domains. If your site is on Cloudflare and no one has opened that setting, assume it's wrong. Test with curl -A "OAI-SearchBot" -I https://yoursite.com. You want a 200.
2. Check the raw HTML. One hour. Many AI crawlers don't execute JavaScript. Run curl https://yoursite.com and read what comes back. If your headline and your services list aren't in it, ChatGPT sees a blank page. Server-side rendering fixes it. Most CMSs already do it. Custom React apps are the usual offender.
3. Verify in Bing Webmaster Tools. 15 minutes. Go to Bing Webmaster Tools. Import your verified properties from Google Search Console in one click. Submit your sitemap. Then open the AI Performance report, in public preview since February 9, 2026. It shows where Copilot and Bing's AI answers cite you; a June 2026 update added citation share and topic groups (details). Copilot runs on the same index as ChatGPT, so this is the closest thing to a ChatGPT citation report anyone will give you.
4. Turn on IndexNow. 30 minutes. IndexNow pushes every URL change to Bing the moment you publish, so nothing waits for a crawl cycle. On Cloudflare it's a toggle under Cache (Crawler Hints). Elsewhere it's a CMS plugin or a hook on publish. A new post can enter ChatGPT's retrieval pool the same day.
5. Open every key page with the answer. Two to four hours. Read the first sentence of your homepage and of your ten most-visited pages. Does it state what the page is, in words a model could quote? "Space is an AI-native brand agency in New York that ships brand and website in sprints of weeks" gets lifted. A hero line about believing in something goes unread. Keep the clever headline; put the plain sentence right under it.
6. Add Q/A blocks with FAQPage schema. Four to eight hours. Models ingest question-and-answer pairs cleanly. Write the questions the way a buyer types them ("how much does a rebrand cost," "what does an AI-native agency do") and answer each in two or three sentences. Mark them up with Schema.org FAQPage. Put them on the homepage and on every service page.
7. Byline, date, updated date. A day of template work. Every long-form page needs a visible author linked to a bio page carrying Person schema with sameAs to LinkedIn, plus a publish date and an updated date. Undated content ranks low in Google and in citation studies alike; the model has no way to tell it's current.
8. Get named where ChatGPT already looks. Ongoing, and the one that matters most. Write down 20 prompts a buyer would type. Run them. Log every source ChatGPT cites. Those sources are your PR list: the listicles, the Reddit threads, the trade publications, the podcasts with show notes. Pitch them. A mention on a page ChatGPT already trusts does more than any change to your own site.
Skip these
There's no way to submit a site to ChatGPT; anyone selling that is selling nothing. ai.txt files go unread. "AI visibility" plugins do what robots.txt and schema do for free. llms.txt is cheap and worth having for coding agents; Google says its Search ignores it. Don't count it as a citation tactic.
How long it takes
Bing indexing takes days once IndexNow is on. Citations take weeks to months; they lag your changes with no dashboard telling you why. Set your 20-prompt list before you touch anything, or you'll never be able to say what worked. [yours: a client timeline, with permission, from unverified in Bing to first citation]
Beacon runs the first seven moves as an audit and returns the fix list in 24 hours, free. The eighth one is on you.
Frequently asked questions
- Does ChatGPT use Bing or Google?
- ChatGPT's live search grounds mostly on Bing's index through OpenAI's Microsoft partnership. OpenAI also runs its own crawler, and some tests show Google's index for paid users. A page absent from Bing is unlikely to be cited.
- Do I need to allow GPTBot to be cited by ChatGPT?
- No. GPTBot collects training data. Citations depend on OAI-SearchBot (the search index) and ChatGPT-User (live fetch). You can block GPTBot and still be cited.
- Can I submit my site to ChatGPT?
- No. There's no submission API. The path runs through Bing Webmaster Tools and IndexNow.
- How long does it take to get cited by ChatGPT?
- Bing indexing takes days with IndexNow. Citations typically lag by weeks to months, depending on domain authority and off-site mentions.
- Does llms.txt help with ChatGPT citations?
- Not measurably. 2026 log studies found most llms.txt files are never fetched by AI retrieval bots. It's useful for coding agents and cheap to ship.
.png%3Fprefix%3Dmedia&w=3840&q=95)