To get your brand cited by ChatGPT, a page has to clear three gates in order: OAI-SearchBot has to be able to fetch it, the words have to be in the HTML it gets back, and the title and first lines have to answer the question well enough to be picked from everything ChatGPT retrieves. You can test all three from a terminal in about ten minutes, and below are the commands, a short Python script, and a real run where an agent did the whole check.
Every guide on this says "allow the crawler" and "answer first". Almost none show you the request, the response, or how to tell whether you passed.
Retrieved is not cited
ChatGPT reads more than it cites. Ahrefs found it pulls in dozens of URLs for a single query and ends up citing about half of them, and that 88% of the URLs it cites come straight from its search index (Ahrefs, read 7 Nov 2026). Their read is that the title, the snippet and the URL make the first call, before ChatGPT opens the page at all.
So there are three filters stacked on top of each other:
- Fetch. Can OAI-SearchBot, ChatGPT's search crawler, get a 200 from your server?
- Read. Is the answer in the HTML it receives, or painted later by JavaScript it never runs?
- Pick. Do the title and the first couple hundred characters answer the question, so the page survives the cut from retrieved to cited?
A failure at gate one makes the other two irrelevant, which is why it goes first.

ChatGPT retrieves dozens of URLs per question and cites about half (Ahrefs, read 7 Nov 2026); each gate is one check you can run.
Gate 1: can OAI-SearchBot fetch the page?
OpenAI documents three agents, and only one of them matters here. OAI-SearchBot is the one that decides whether your site can appear in ChatGPT's search answers. GPTBot crawls for model training. ChatGPT-User fetches a page when a person asks ChatGPT about it and plays no part in search inclusion (OpenAI's crawler docs, read 7 Nov 2026). The search and training settings are independent, so you can stay citable and still opt out of training.
Step 1: read your robots.txt the way a crawler does
This is the part nobody mentions. A crawler doesn't merge groups. It follows the one group that names it most specifically: its own User-agent: line if there is one, else User-agent: *. Inside that group the longest matching path wins, and Allow wins a tie.
That turns a well-meant fix into a leak. This file looks like it only adds ChatGPT:
User-agent: * Disallow: /admin/ Disallow: /drafts/ User-agent: OAI-SearchBot Allow: / OAI-SearchBot now ignores the * group entirely, so /admin/ and /drafts/ are open to it. Repeat the rules you care about inside its own group:
User-agent: * Disallow: /admin/ Disallow: /drafts/ # ChatGPT search: citable User-agent: OAI-SearchBot Disallow: /admin/ Disallow: /drafts/ Allow: / # OpenAI training: opted out User-agent: GPTBot Disallow: / That group rule is spelled out in the crawler playbook supereo's MCP server returned on 4 Nov 2026 (more on that run below).

A crawler follows only the most specific group that names it, so rules under User-agent: * stop applying once it has its own.
A quick check with the standard library:
from urllib.robotparser import RobotFileParser rp = RobotFileParser("https://example.com/robots.txt") rp.read() for bot in ("OAI-SearchBot", "GPTBot", "ChatGPT-User"): print(f"{bot:14} {rp.can_fetch(bot, 'https://example.com/pricing')}") One catch: urllib.robotparser walks a group's rules in file order and stops at the first match, where crawlers use the longest match. Put the specific Disallow lines above the broad Allow: /, as in the file above, and both readings agree.
OpenAI's docs say a robots.txt change takes about 24 hours to reach ChatGPT search. Plan your test for the next day.
Step 2: check what your CDN sends that name
OpenAI's eligibility steps have a second half. Your host or CDN has to let traffic from OpenAI's published search-bot IP addresses through (OpenAI help, read 7 Nov 2026). robots.txt can say yes while a bot rule in front of it says no.
Ask your server what it sends the crawler's user agent, then what it sends a plain request:
UA='Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.4; +https://openai.com/searchbot' curl -s -o /dev/null -w 'as OAI-SearchBot: %{http_code}, %{size_download} bytes\n' -A "$UA" https://example.com/pricing curl -s -o /dev/null -w 'plain request: %{http_code}, %{size_download} bytes\n' https://example.com/pricing That user agent string is the exact one supereo's fetch_as_bot tool sent on 7 Nov 2026. A 403, a 503, or a body a fraction of the size of the plain one means something in front of your app treats the name as a threat.
The limit: this request comes from your address, not OpenAI's. A firewall that checks IPs can wave you through and still block the real crawler. Your logs are the only place that shows it:
grep 'OAI-SearchBot' /var/log/nginx/access.log | awk '{print $9}' | sort | uniq -c You want a column of 200s. No lines at all over a couple of weeks means it isn't reaching you. On Cloudflare, open the AI crawler settings too. Cloudflare moved to blocking AI crawlers by default in 2025, then changed the defaults for new domains again in September 2026 (Cloudflare docs, read 7 Nov 2026), so what your zone does depends on when it joined and what someone clicked back then.
Gate 2: is the answer in the HTML it gets?
Vercel looked at AI crawler traffic and found ChatGPT's and Claude's crawlers download JavaScript files without executing them, so anything rendered in the browser is invisible to them (Vercel, read 7 Nov 2026). A React app that ships an empty root div and paints the pricing table after hydration has no pricing table as far as ChatGPT is concerned.
curl | grep tells you whether one string is there. This script tells you what a non-JS crawler actually gets: the status, the title, how many words of real text, the first 200 characters after the H1, how many of your question's words land in them, and the JSON-LD types with the Organization's sameAs.
# check_page.py: what a crawler that doesn't run JavaScript gets from one URL. # usage: python3 check_page.py <url> "<the question the page answers>" import json import re import sys import urllib.error import urllib.request from html.parser import HTMLParser UA = ("Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 " "(KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; " "OAI-SearchBot/1.4; +https://openai.com/searchbot") HIDDEN = {"script", "style", "noscript", "template", "svg"} class Page(HTMLParser): def __init__(self): super().__init__() self.hidden = 0 # depth inside tags a reader never sees self.where = None # "title", "h1" or "ld" while inside one self.title, self.text, self.ld = "", [], [] self.h1_end = None # where the first H1 stops, as an index into text def handle_starttag(self, tag, attrs): if tag == "script" and dict(attrs).get("type") == "application/ld+json": self.where = "ld" self.ld.append("") elif tag in HIDDEN: self.hidden += 1 elif tag in ("title", "h1") and self.where is None: self.where = tag def handle_endtag(self, tag): if tag == "script" and self.where == "ld": self.where = None elif tag in HIDDEN: self.hidden = max(0, self.hidden - 1) elif tag == self.where: self.where = None if tag == "h1" and self.h1_end is None: self.h1_end = len(self.text) def handle_data(self, data): if self.where == "ld": self.ld[-1] += data elif self.where == "title": self.title += data elif not self.hidden and data.strip(): self.text.append(data.strip()) def walk(node): if isinstance(node, list): for x in node: yield from walk(x) elif isinstance(node, dict): yield node for v in node.values(): yield from walk(v) def schema(blocks): types, same_as = set(), [] for block in blocks: try: nodes = list(walk(json.loads(block))) except ValueError: continue for node in nodes: kind = node.get("@type", []) kind = kind if isinstance(kind, list) else [kind] types.update(k for k in kind if isinstance(k, str)) if "Organization" in kind: links = node.get("sameAs", []) same_as += links if isinstance(links, list) else [links] return sorted(types), [str(s) for s in same_as] def main(url, question): req = urllib.request.Request(url, headers={"User-Agent": UA}) try: with urllib.request.urlopen(req, timeout=20) as r: status, html = r.status, r.read().decode("utf-8", "replace") except urllib.error.HTTPError as e: sys.exit(f"status {e.code}: the crawler gets nothing to read") page = Page() page.feed(html) words = re.findall(r"[A-Za-z0-9][A-Za-z0-9'-]*", " ".join(page.text)) opening = " ".join(page.text[page.h1_end or 0:])[:200] asked = [w for w in re.findall(r"[a-z0-9]+", question.lower()) if len(w) > 3] found = [w for w in asked if w in opening.lower()] types, same_as = schema(page.ld) print(f"status {status}") print(f"title {page.title.strip()}") print(f"words {len(words)}") print(f"opening {opening}") print(f"question {len(found)} of {len(asked)} words in the opening: {found}") print(f"json-ld {', '.join(types) or 'none'}") print(f"sameAs {', '.join(same_as) or 'EMPTY'}") if __name__ == "__main__": main(sys.argv[1], sys.argv[2]) Run it on the page you want cited, with the question a buyer would type:
python3 check_page.py https://example.com/pricing "how much does example cost per seat" What each line should look like:
| Line | Pass | Fail, and the usual cause |
|---|---|---|
| status | 200 | 403 or 503: a bot rule at the CDN (gate 1) |
| words | Hundreds, close to what the browser shows | Under 50: client-side rendering |
| opening | States the answer and names the product | A hero tagline, cookie text or nav links |
| question | Most of the question's words | 0 or 1: the page answers a different question |
| json-ld | Organization plus the page's own type (Article, FAQPage, Product) | Empty, or injected by a tag manager after load |
| sameAs | Your real profiles | EMPTY: nothing outside your site vouches for the name |
Gate 3: do the title and first 200 characters answer it?
Being fetchable and readable gets you into the pile. Being picked is the next filter. Ahrefs puts the title, snippet and URL ahead of the content in that decision (Ahrefs), and found ChatGPT searches several variations of a question, not only the one typed (Ahrefs, read 7 Nov 2026).
Search Engine Land published an audit of blog posts on 15 domains that drew ChatGPT referral traffic. 72.4% of the cited posts had an answer capsule, and about 91% of those capsules held no links at all (Search Engine Land, read 15 Nov 2026).
The guides disagree on how long that answer should be. Most say 40 to 60 words inside the first 100. The SEL audit defined a capsule as 120 to 150 characters, about 20 to 25 words, right under a question-shaped title or H2. I'd write to the stricter one: the answer in the first 200 characters under the H1, then the longer version right after it. That window is what the script's opening line prints.
The edits that usually fix gate 3:
- The title uses the question's words. "Pricing" becomes "Example pricing: plans per seat, 2026".
- The first sentence is the answer, with your product's name in it and no links.
- The URL is plain words, like
/pricingor/guides/how-to-get-cited-by-chatgpt, never/p?id=4471.
A real run: Claude Code checks a page over MCP
Doing this page by page gets old fast, so here's the same check run by an agent. The session ran Claude Code 2.1.299 on claude-opus-5-5, started with --bare and no built-in tools, with supereo's MCP server (its staging environment) as the only tool. supereo (supereo.app) is an AI agent that does everything it takes to get your brand recommended by ChatGPT, Gemini, Perplexity, Claude and Google, on your site and off it: site fixes, new pages, roundups, placements and video. It proves what worked against control questions. Made by superagnt.
The prompt asked it to check whether AI crawlers can read supereo.app, look at the About page the way ChatGPT's search crawler sees it, and say what to fix first, briefly. On 4 Nov 2026, Claude Code picked three calls:
| Call | Round trip | What came back |
|---|---|---|
| check_ai_access | 360 ms | robots.txt read; OAI-SearchBot, GPTBot, ChatGPT-User and seven other crawlers allowed; llms.txt served |
| fetch_as_bot, as OAI-SearchBot | 190 ms | 200 in 28 ms, 38,005 characters of HTML, no redirects |
| audit_page | 239 ms | 480 words, no JS shell, one H1, six H2s (five phrased as questions), dated 4 Nov 2026, no noindex |
An excerpt of audit_page's signals, with the URL fields left out:
{ "h1Count": 1, "words": 480, "jsShell": false, "questionHeadings": 5, "structuredData": ["Organization", "Person", "SoftwareApplication", "Offer", "UnitPriceSpecification", "AboutPage", "WebSite"], "noindex": false, "aiCrawlersBlocked": [], "llmsTxtServed": true } Gates 1 and 2, passed. Then the reply flagged what none of the checks had:
Fix first: the brand's own
"sameAs": []is empty in both the Organization and SoftwareApplication data. Only the founder has profiles linked. The page's main job is to tell supereo apart from SuperEA, Superego and Supereo LTD, and right now nothing outside supereo.app backs that up.
That's an entity gap, and no crawler check catches it. The page is readable, and it still gives ChatGPT no outside profile to tie the name to, which hurts most when your name looks like three other companies' names. The reply also noted its own limit: the fetch came from supereo's address, so it can't rule out a block on the real crawler at the edge. Same limit as the curl test.
A second pass the same day with get_page_diagnosis scored the page 84 and failed one check, "Real text in the HTML (cited homepages carry about 900 words)", with the page at 480. Its read of the content said the page answers up front only "partly" and cites no sources. So the page cleared gates 1 and 2, half cleared gate 3, and had an entity gap on top. The sourced version of every step here, with a different recorded run, is in supereo's guide on how to get cited by ChatGPT.

Claude Code with supereo's MCP server, 4 Nov 2026: check_ai_access, fetch_as_bot, audit_page, then get_page_diagnosis on the same page.
Off the page: what ChatGPT reads about you
The three gates decide whether your own page can be cited. For category questions ("best X for Y"), the cited page is often someone else's. Ahrefs found recently updated "best X" lists were the most prominent page type in ChatGPT's sources for top-of-funnel queries (Ahrefs, read 7 Nov 2026), and that the URLs ChatGPT cites run 458 days newer than Google's organic results (Ahrefs, read 7 Nov 2026).
So two kinds of pages carry the brand. Your docs, help, pricing and home pages answer the questions about you, and they need a visible date that moves when the facts do. The lists, reviews and comparisons answer the questions about your category, and you need to be on them. A citation also isn't a recommendation: ChatGPT can cite your pricing page and still recommend a competitor, which is why the same one-sentence description on your site, your profiles and those lists matters as much as any crawler rule.
How to tell if it worked
- ChatGPT adds
utm_source=chatgpt.comto the referral links it sends you (OpenAI publishers FAQ, read 15 Nov 2026). Filter your analytics on it. - Ask the same buyer questions in ChatGPT with search on, on a schedule, and log which URLs it cites. Answers vary run to run, so one hit is noise.
- Keep a few questions your change couldn't affect and ask those too. If they move as much as your targets, the cause was outside your page: a model update, a competitor's launch, a slow week. That's the control-question setup supereo grades its own changes against.
Short answers
Can I pay to get cited by ChatGPT? No. OpenAI says ads run on separate systems from the chat model, and advertisers can't shape, rank or alter its answers (OpenAI help, read 7 Nov 2026).
Does blocking GPTBot stop ChatGPT from citing me? No. GPTBot is the training crawler. Citations ride on OAI-SearchBot, and you can block one and allow the other.
Does schema get me cited? It tells ChatGPT who you are and what the page is, and it only counts if it's in the server-rendered HTML. ChatGPT's own answer to this question says schema doesn't guarantee a citation.
How do I see what ChatGPT cited? In an answer that used search, each citation opens its source, and on desktop you can hover one to preview it (OpenAI help).
How do I cite ChatGPT in APA 7? APA now asks you to cite the specific chat rather than the tool: the company, the date of the chat, the chat's title in italics, a description such as Generative AI chat, the tool and model, then the chat's URL (SNHU library, read 7 Nov 2026).
How do I properly cite a website, or get the citation details from one? You need the author, the date, the page title, the site name and the URL. Put all five in your own pages' visible text and in their Article JSON-LD. A page that says who wrote it and when is easier for a person to cite, and it hands ChatGPT the date it weighs.

