Getting cited by ChatGPT or by Google's AI answers is not a switch you flip in a dashboard. Three things have to line up: a crawler can read the page, one passage clearly answers a question, and the page gives the model verifiable details — a date, an author, a source, a number. Special files, micro-chunking tricks and bought mentions do little or nothing on their own.
The audience is real. Google reported 2.5 billion monthly active users for AI Overviews and more than one billion for AI Mode in a post published June 3, 2026 and updated August 31, 2026. Those answers link to pages that help them respond. Your job is to be one of those pages, not to guess a hidden signal.
This tutorial takes about 45 minutes on a site that is already indexed. You will check crawler access, add Article markup, rewrite a few passages, submit URLs with IndexNow and set up honest measurement. The commands are ready to adapt to your domain.
What GEO means in practice#
GEO stands for generative engine optimization. The label covers work aimed at AI answers: Google's AI Overviews and AI Mode, ChatGPT Search, and Perplexity. It is not a separate indexing system.
Google's position is blunt. Its AI features documentation says there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." Its optimization guide adds that you do not need new machine-readable files, AI text files or special schema markup. For Google Search, GEO is SEO. Anyone selling it as a separate magic layer is skipping that sentence.
The retrieval mechanics matter more than the acronym. Google describes two techniques. Retrieval-augmented generation, or RAG, pulls relevant pages from the Search index and uses their text to ground a response. Query fan-out sends several related searches at once to cover subtopics. A question about the best hosting for a solo developer can trigger searches about price, RAM limits, backups and support.
The practical result: a page has to be indexable, organized into readable sections, and contain self-contained answers. It also has to cover the follow-up questions a reader would ask, without publishing one thin page per query variation.
ChatGPT and Perplexity do not read Google's index. They run their own crawlers and publish their own rules. The table below summarizes what each agent controls, based on the official documentation checked in November 2026.
| Crawler | What it does | robots.txt token | Delay or limit |
|---|---|---|---|
| Googlebot | Google Search, including AI Overviews and AI Mode | Googlebot | Recrawl from several days to several months |
| Google-Extended | Gemini training and grounding in Gemini apps, not Search | Google-Extended | Does not affect inclusion in Search |
| OAI-SearchBot | ChatGPT search results and citations | OAI-SearchBot | About 24 hours after a robots.txt change |
| GPTBot | OpenAI model training | GPTBot | About 24 hours |
| ChatGPT-User | Fetches a page when a user asks | ChatGPT-User | robots.txt may not apply |
| PerplexityBot | Perplexity index and results | PerplexityBot | Up to 24 hours |
| Perplexity-User | Fetches a page when a user asks | Perplexity-User | Generally ignores robots.txt |
The most misunderstood row is GPTBot. Blocking GPTBot keeps your content out of OpenAI's training data. It does not remove you from ChatGPT search answers; OAI-SearchBot handles that. OpenAI says so directly: a site opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though it can still appear as a navigational link.
1. Check that AI crawlers can read your pages#
Start with the simplest file on the site. A correct robots.txt does not guarantee a citation, but a file that blocks everything makes the rest irrelevant. This configuration allows search, blocks GPTBot training and lets the AI search crawlers through:
User-agent: *
Allow: /
User-agent: GPTBot
Disallow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
Sitemap: https://example.com/sitemap.xmlCrawlers with their own group read that group, not the general rule. If you also want to allow training, remove the GPTBot block. That is your call, but make it deliberately. Note that Google-Extended does not control AI Overviews. Search access is controlled through Googlebot, as Google's AI features documentation states.
Then test the wall, not just the file. A permissive robots.txt is useless if a CDN, firewall or anti-bot rule returns a challenge page to the crawler. These commands send the main crawler user agents and print the HTTP status:
for bot in "Googlebot/2.1 (+http://www.google.com/bot.html)" "OAI-SearchBot/1.4" "PerplexityBot/1.0"; do
printf "%-45s " "$bot"
curl -sS -o /dev/null -w "%{http_code}\n" -A "$bot" https://example.com/your-page
doneA 200 means that user-agent string was not blocked. It does not prove the crawler will get through: a WAF can challenge it after a few requests, and user agents can be spoofed. To go further, compare request IPs with the ranges published by OpenAI (openai.com/searchbot.json, openai.com/gptbot.json) and Perplexity (perplexity.ai/perplexitybot.json). Those files are the source of truth for firewall rules.
Cloudflare is the classic trap. Perplexity's documentation says Bot Fight Mode cannot be skipped with a custom WAF rule: it can still challenge crawlers even after you create an allow rule. If your site sits behind Cloudflare and your logs never show OAI-SearchBot, check that setting first. Our Next.js hosting comparison with Cloudflare covers the reverse-proxy and firewall details worth reviewing.
One more check: if the page is behind noindex, a login or a consent screen that hides the content, the crawler will not see the answer. Google can render JavaScript, but the HTML served at the start is still the safest path. Put the main content in HTML, not in a component that appears after three clicks.
2. Add signals a model can verify#
Markup does not create a citation. It removes ambiguity. Google repeats that no special schema is required for AI features, but Article markup still identifies the headline, author, publication date and update date. Those details help a system decide whether a page is fresh enough to answer a question about a price or a version.
Here is a JSON-LD example for the <head>. Replace the URLs and dates with yours:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "How to get cited by ChatGPT and Google AI answers",
"description": "GEO without the jargon: crawlers, markup, quotable passages and measurement.",
"datePublished": "2026-11-06T09:00:00-05:00",
"dateModified": "2026-11-06T09:00:00-05:00",
"author": {
"@type": "Person",
"name": "Terra",
"url": "https://example.com/about"
},
"publisher": {
"@type": "Organization",
"name": "Le Dev Augmenté",
"url": "https://example.com"
},
"mainEntityOfPage": {
"@type": "WebPage",
"@id": "https://example.com/your-page"
},
"image": [
"https://example.com/images/blog/your-page.webp"
]
}
</script>The fields that earn their keep are datePublished, dateModified, author.url, publisher and mainEntityOfPage. Google recommends linking the author to a page that identifies them, not just printing a name. Its Article documentation also says the markup must match visible text: do not declare an update date the page does not show.
FAQ markup needs an update. Google removed the FAQ rich result from Search on May 7, 2026. Keeping FAQ markup is not forbidden, but it will no longer produce the expandable result. Keep a visible FAQ in HTML for readers and for models that read the page; do not build a strategy around a rich result that no longer exists.
Validate the JSON-LD with a script, not by eye. This one fetches the page, extracts every application/ld+json block and prints the type, headline and modification date:
python3 - <<'PY'
import json, re, urllib.request
url = "https://example.com/your-page"
html = urllib.request.urlopen(url, timeout=10).read().decode("utf-8")
blocks = re.findall(r'<script type="application/ld\+json">(.*?)</script>', html, re.S)
for i, raw in enumerate(blocks, 1):
data = json.loads(raw)
print(i, data.get("@type"), data.get("headline"), data.get("dateModified"))
PYIf the script throws a parsing error, the block is invalid and Google will ignore it. Fix the comma or the quote before adding more fields.
3. Write passages an AI answer can quote#
This is where most of the work lives, and it survives algorithm changes better than any trick. An AI answer is built from passages it can cite. A quotable passage carries one idea, answers a question and includes enough context to make sense outside the page.
The working rule: an H2 asks a real question, the first paragraph answers it in 40 to 80 words, and the next paragraphs add conditions, exceptions and numbers. Here is the shape.
How long does Google take to index a page? Google can take several days to several months to crawl and index a page. The delay depends on how often the site changes, internal links and page quality. Google does not guarantee either the delay or indexing.
That passage is short, but it answers. It does not say "it depends" without a frame, and it does not promise a position. A model can reuse it with the nuance intact.
Compare it with copy that gives a model nothing to quote:
| Avoid | Write instead |
|---|---|
| "In today's fast-moving world of SEO, indexing is a complex topic." | "Google crawls and indexes a page in anywhere from several days to several months. It does not guarantee indexing." |
| "Our revolutionary tool optimizes your AI visibility." | "The tool checks the HTTP codes returned to crawlers and flags blocked pages." |
| "Many experts believe that…" | "OpenAI's documentation says a robots.txt change takes about 24 hours to apply." |
| "The best price on the market." | "ChatGPT Plus costs $20/month, a price checked in November 2026." |
Models cite pages that give them something checkable: a tested version, a date, a machine, a price paid, a sample, a limit hit. Add those details to tutorials and comparisons. A test dated November 6, 2026 on an RTX 4090 with 24 GB of VRAM is worth more than a paragraph that repeats a product page.
Do not publish one page per query variation. Google calls scaled production of pages whose main purpose is to manipulate rankings or AI responses "scaled content abuse." Query fan-out does not ask you to duplicate content; it asks you to cover the follow-up questions in the same page, with distinct sections.
Internal links still matter. A page about GEO should link to your tutorials, comparisons and definitions. Those links help crawlers understand context and help readers go deeper. Avoid generic anchors such as "click here"; write the topic of the link.
Keep the main content in HTML. Google can render JavaScript, but models reading a page to answer a question do not always have the time or the resources to run a full application. Content that appears after an asynchronous load can be invisible to some agents.
4. Get indexed faster#
An up-to-date sitemap does not force indexing, but it helps engines discover new URLs. Declare each URL once, with an honest lastmod, and submit the sitemap in Search Console. Do not change lastmod without a real edit; engines learn to ignore it.
For Bing and the engines that participate in IndexNow, you can submit a URL in seconds. The key must contain 8 to 128 hexadecimal characters, and the verification file must be served at the domain root. One POST accepts up to 10,000 URLs.
KEY="replace-with-8-to-128-hexadecimal-characters"
printf '%s' "$KEY" > "/var/www/example.com/${KEY}.txt"
curl -sS -X POST https://api.indexnow.org/indexnow \
-H "Content-Type: application/json; charset=utf-8" \
-d "{\"host\":\"example.com\",\"key\":\"${KEY}\",\"urlList\":[\"https://example.com/your-page\"]}"A 200 means the engine received the URL, not that it indexed it. Google does not use IndexNow: for Google, use the sitemap, URL Inspection and, when needed, a recrawl request. ChatGPT and Perplexity do not offer a public URL submission endpoint. They crawl with their own bots, and a robots.txt change can take about 24 hours to apply.
Indexing speed is not a citation lever. An indexed page can go uncited; an unindexed page cannot be cited. Treat indexing as a prerequisite, not a goal.
5. Measure citations and visits#
Google provides a dedicated report in Search Console: the generative AI performance report. It shows impressions in AI Overviews and AI Mode, broken down by page, country and device. It rolled out to all sites on August 31, 2026. Clicks remain in the main Performance report, under the "Web" search type.
ChatGPT and Perplexity have no public equivalent. Measure with two sources: server logs and a manually tested prompt panel.
For logs, first count crawler visits. This command shows crawl frequency, not citations:
grep -Eo "(Googlebot|OAI-SearchBot|PerplexityBot|GPTBot)/[0-9.]+" /var/log/nginx/access.log \
| sort | uniq -c | sort -nrThen look for referrers. AI clients often strip the header, but some clicks still arrive with a source URL:
grep -Eo "https?://(chatgpt\.com|perplexity\.ai|gemini\.google\.com)[^\" ]*" /var/log/nginx/access.log \
| sort | uniq -c | sort -nr | head -20A missing referrer does not mean you were not cited. Many ChatGPT visits arrive as direct traffic. Cross-check logs with brand search volume and direct visits to the pages that answer your target questions.
The prompt panel is more reliable for citations. Pick 20 real questions your audience asks ChatGPT, Perplexity and Google AI Mode. Ask them on the same day, in the same language, in a private window, and record the cited domain, cited URL, date and wording. Repeat once a month. Twenty questions are not a statistical study, but they reveal which pages enter answers and which disappear.
For classic SEO tracking, our open source Semrush and Ahrefs alternatives covers tools for positions, backlinks and queries. None of them sees inside Google, ChatGPT or Perplexity. They measure traffic and rankings, not citations.
What does not work, or barely does#
The llms.txt file. Jeremy Howard proposed the idea in September 2024 and revised it in August 2026. The file is a Markdown overview that points agents to useful pages. It helps technical documentation read by agents. Google says Search ignores llms.txt and that the file neither helps nor hurts visibility. Do not sell it as a citation lever.
Micro-chunking. Google says there is no requirement to break content into tiny pieces for AI to understand it. Clear sections help readers and give retrieval systems clean passages. That does not mean chopping every sentence.
FAQ markup for a rich result. The FAQ rich result has not been shown since May 7, 2026. A visible FAQ is still useful; the markup will not bring the expandable display back.
Bought mentions and fake reviews. Google says inauthentic mentions are not a reliable strategy, and its spam policies apply to generated answers too. An undisclosed paid mention can cost more than it returns.
Tools that claim access to Google's "internal metrics." No third-party tool has access to Google's ranking or AI systems. A tool can help you track prompts or positions; it cannot sell you a place in an answer.
A ChatGPT Plus subscription. Prices checked in November 2026 put ChatGPT Plus at $20/month. That buys models and usage limits, not a spot in answers. OAI-SearchBot does not read your subscription.
The 8-point checklist#
- Googlebot is not blocked, the page returns
200and has nonoindex. - OAI-SearchBot and PerplexityBot are allowed if you want citations.
- The GPTBot block is a documented choice, not an oversight.
- Your CDN and WAF let verified crawlers through; Bot Fight Mode is tested.
- The
ArticleJSON-LD is valid and matches visible text. - Each H2 answers a real question, with a 40-to-80-word answer up top.
- Every number has a date, a version or a source.
- The sitemap is current, IndexNow is wired, and the 20-prompt panel runs monthly.
Frequently asked questions
Do I need an llms.txt file to be cited by ChatGPT?
No. llms.txt is an open proposal to help agents find documentation. Google Search says it ignores the file and that it has no effect on visibility. ChatGPT and Perplexity publish no rule making llms.txt a condition for citation. The file can still help your docs; it is not a GEO lever.
How long does it take to appear in ChatGPT or AI Overviews?
There is no guaranteed delay. Google can take several days to several months to crawl and index a page. OpenAI says a robots.txt change takes about 24 hours to apply, and Perplexity gives a delay of up to 24 hours. Those delays cover rule changes, not the citation itself.
Does blocking GPTBot remove me from ChatGPT answers?
No. GPTBot is for model training. OAI-SearchBot controls whether your site appears in ChatGPT search results. If you block OAI-SearchBot, your site will not be shown in search answers, though it can still appear as a navigational link. You can block training and allow search.
Is FAQ markup still useful for citations?
Google removed the FAQ rich result on May 7, 2026. The markup no longer produces the expandable display. A visible HTML FAQ remains useful for readers and for models that read the page, but do not build a strategy around that rich result.
How do I know whether ChatGPT cites me?
There is no official ChatGPT citation report. Use logs to spot OAI-SearchBot, referrers for the clicks that keep one, and a 20-question panel tested monthly in ChatGPT, Perplexity and Google AI Mode. Record the cited domain and URL, not just traffic.
Can I buy mentions to get cited faster?
No. Google says inauthentic mentions are not a reliable strategy and that its spam policies apply to generated answers. A mention should be real, dated and verifiable. Original content, primary sources and a readable page remain the work that pays.
Sources#
Pages and documentation checked on November 6, 2026.
- Google Search Central — AI features and your website
- Google Search Central — Optimizing your website for generative AI features
- Google — New opportunities, control and insights for website owners
- Search Console — Search generative AI control
- Search Console — Generative AI performance report
- Google Search Central — Common crawlers and Google-Extended
- Google Search Central — Article structured data
- OpenAI — Overview of OpenAI Crawlers
- Perplexity — Perplexity Crawlers
- IndexNow — Documentation
- llms.txt — The /llms.txt file, v2
- OpenAI Help — What is ChatGPT Plus?



Comments
Loading comments…