How to get your content cited by AI

Ask ChatGPT or Google's AI a question and it writes you an answer, then lists a few websites it took that answer from. Those named sites get the click and the trust. Getting your page onto that short list is the goal, and there is a whole practice built around it. Here is what works, backed by a study that measured it, and a prompt that checks your page from top to bottom.

The four changes a study measured

In 2023 a Princeton team ran a controlled test of about ten thousand queries and ranked which content changes made AI answers cite a page more often. Their results give you real levers instead of guesswork:

• Quote a named source. Adding a quotation from a credible source lifted citations by about 41%, the biggest single move.
• Add a real statistic. Working in genuine numbers raised citations by around 33%, and it held up across every topic they tested.
• Cite your sources. Showing where your claims come from added roughly 28%, because it makes the page quotable and checkable.
• Write cleanly. Plain, fluent writing added about 28% on its own. A page the AI can read easily is a page it can lift from easily.

The one that backfired is the one old SEO taught you: keyword stuffing lowered citations by about 8%. Packing the page with the phrase you want to rank for now works against you, so the habit to break is stuffing, and the habit to build is quoting, counting and citing.

Here is the shape on a single sentence, with the brackets yours to fill with something real:

• Before: "Our AI notetaker is the best AI notetaker for busy teams who want the best notes."
• After: "The best free AI notetaker for most people is [tool]. A [year] test by [source] clocked it transcribing a one-hour call in [X] minutes, and its founder says accuracy comes from [short quote]."

Same claim, but the after version leads with a direct answer, carries a real number and a named quote, and the keyword padding is gone.

What not to waste time on

Plenty of GEO advice is invented. Google's own guidance says there is no special AI optimisation: the same people-first SEO that earns normal search visibility is what makes you eligible for AI answers. It goes further and tells you what to skip, no made-up llms.txt file, no artificially chopping your content into chunks, no chasing fake brand mentions. Be genuinely useful, be crawlable, use clear headings, and let the four levers above do the lifting.

The Substack trap

A lot of creators assume that publishing on Substack gets them found by AI. Michael Brito analysed 1.7 million prompts and found Substack named in just 0.07% of AI answers, against Medium's 0.36%, roughly five times fewer. The reason is plain: Substack's paywall blocks the crawlers that feed the engines. Keep Substack for building an audience, and if you want AI to quote your thinking, put it somewhere crawlable as well.

First, check a crawler can read it

Before you tune anything, make sure an engine can reach the page at all. If it sits behind a login, a paywall or a members-only gate, the crawlers that feed AI answers cannot read it, the same reason Substack barely shows up. Open the page in a private browser window while logged out: if you can read the whole thing, so can a crawler. Then search Google for site:yourdomain.com, swapping in your own address. If your page does not come back, you are not indexed yet, and none of the levers above will help until that is sorted.

Check whether AI names you

Before you pay for anything, do the free version. Open ChatGPT in search mode, Perplexity, and Google's AI answers, and ask the exact questions a customer would ask in your niche. See whether you are named or linked, and note which questions surface you and which do not. Do it once a month so you can see it move. If you want it tracked over time, a paid tool like Otterly.ai watches ChatGPT, Perplexity and Google's AI answers for you and has a free trial to test it first.

The prompt that audits your whole page

This covers the whole page, the writing and everything around it. It checks whether AI can reach the page, whether it is built so machines can read it, and whether the content is quotable, then it fixes what it can see, hands you a plain list for the server-side bits it cannot, and gives you weekly prompts to test whether you are getting cited. It is honest that none of this guarantees a citation, and it never invents a number or a quote to hit a target.

You are my citability auditor and editor. Your job is to check everything that decides whether an AI answer tool can find, read and quote my page, then fix what you can see in the paste and give me a plain list of the rest. Getting cited is worthless if the page lies, so you never invent a fact, a number or a quote to hit a target.

Honest scope: following this improves eligibility and quotability. It does not guarantee a citation. Time, competition and off-site trust still matter. There is no public "submit URL to get cited" button for ChatGPT or Perplexity for a typical site.

SET UP (fill this in)
- THE PAGE: [paste the full page text, and the URL if it is live]
- THE QUESTION IT SHOULD ANSWER: [the exact question a reader would ask an AI, for example "what is the best free AI notetaker"]
- WHO ASKS IT: [the reader behind the question, for example a beginner choosing their first tool]
- MY REAL MATERIAL: [paste any genuine statistics, quotes or sources I can use; leave blank if I have none]
- OPTIONAL SITE NOTES: [paste robots.txt, a sitemap excerpt, Search Console or Bing notes if you have them; leave blank if not]

BEFORE YOU START
If THE QUESTION is missing or too broad to answer in two sentences, ask me for it and wait. Say plainly that you can only see what I paste, so for anything on the server (robots.txt, the sitemap, HTTPS, Search Console, Bing, IndexNow, page speed) you will tell me how to check it rather than pretend you can see it.

PART 1: CAN AI REACH THE PAGE
Mark each Pass, Fix, or Check yourself, with the fix in one line:
- Loads over HTTPS, not plain HTTP.
- Not blocked in robots.txt, and has no "noindex" tag on the live page.
- Listed in your sitemap.xml.
- Has one canonical tag pointing to itself, so duplicates do not split it.
- Not behind a login or paywall, so a crawler can read the whole thing.
- Works on mobile and loads quickly, with images compressed.
- No broken links.
For anything you cannot see in what I pasted, tell me exactly where to look.

PART 1B: DISCOVERY BEYOND THE PAGE FILE (Check yourself only unless I pasted proof)
Mark each Pass, Missing, or Check yourself. Do not invent that these are done.
- Google Search Console: site verified; sitemap submitted; for important new/updated URLs use URL Inspection then Request indexing; respect the daily quota; if discovered URLs lag your live sitemap, resubmit or refresh the sitemap.
- IndexNow: a key file is live on your host; when you publish or update important URLs you notify IndexNow (for example POST to https://api.indexnow.org/indexnow). IndexNow notifies participating engines; it does not guarantee crawl, index or citation.
- Bing Webmaster Tools (optional but useful): site added or imported; same sitemap submitted so you can see Bing discovery separately from Google.
- robots.txt: allows normal crawling unless you deliberately block a bot; includes a Sitemap line.
- llms.txt: Google advises against making one, so skip it and keep your sitemap and robots.txt clean instead.
- Do not waste time "submitting" the URL inside ChatGPT or Perplexity chat as if that were Search Console. That is not a citation API.

PART 2: IS IT BUILT SO MACHINES CAN READ IT
Mark each Pass or Fix, with the fix:
- Exactly one H1, then a logical order of H2 and H3 headings under it.
- A clear meta title and meta description that match the question.
- A short, readable URL made of real words.
- Alt text on every image.
- An OG image set, so it looks right when shared.
- Internal links from this page to related ones, and from your other pages back to it.
- Schema markup where it fits, for example FAQ markup on a page of questions, and only if the visible FAQ text matches the schema word-for-word.

PART 3: IS THE CONTENT QUOTABLE
Rewrite the content so it does these, using only what is real:
- LEAD WITH THE ANSWER. Put a direct, plain answer to the question in the first two sentences. On a product or service page, that is what it is and who it is for, before any selling.
- ADD ONE REAL STATISTIC. If I have not given you one, mark [NEED STAT] and tell me what to find. Never invent a number.
- ADD ONE NAMED-SOURCE QUOTE. If I have not supplied one, mark [NEED QUOTE]. Never invent a quote.
- SHOW YOUR SOURCES for every claim, so the page is quotable and checkable. Prefer first-party docs for product facts.
- CUT THE PADDING. Remove keyword stuffing, say each thing once, and rewrite headings as the questions a reader would ask.
- ONE JOB. Keep the page on this question; do not turn it into a vague mega-pillar.

RULES
- Never invent a statistic, quote, source, date, traffic lift or "already cited by X" claim. Flag gaps with [NEED STAT] or [NEED QUOTE].
- Keep my meaning and my voice.
- For anything you cannot verify from the paste, say how I check it rather than guessing.

THEN GIVE ME
1. A priority list of fixes, worst problems first, each with its one-line fix.
2. The rewritten content from Part 3.
3. The short list of things only I can do (server-side and console items), spelled out so I can hand them to whoever runs the site.
4. Five weekly test prompts (tied to THE QUESTION) to check in Perplexity and, when shown, Google AI Overviews, to see if my domain is cited. Remind me that no citation yet is normal for a while after indexing.

SELF-CHECK
- Show me, in one line, the direct answer you put at the top.
- List every number and quote you used and where each came from, plus everything marked [NEED STAT], [NEED QUOTE] or Check yourself.
- Name the single change most likely to get this page cited, and say honestly if the blocker is content, indexing or something you cannot see.

Do one now

Pick the one page you most want AI to quote, the one that answers a question your customers actually ask. Run the prompt on it, fill in the real stat and quote it flags, and publish the tighter version. Then ask ChatGPT and Perplexity that question next week and see if you have moved. If the writing itself needs sharpening first, the AI writing tells to cut will help it read cleanly, which is a lever in its own right.

A few quick questions

Is this different from normal SEO?

The foundations are the same: your page has to be crawlable, indexed, and clearly written. The emphasis shifts. A Princeton study found that quoting a named source, adding a real statistic and writing cleanly raised how often AI cited a page, while keyword stuffing lowered it by 8%. Google says there is no separate AI optimisation, so do people-first SEO and add those citable signals on top.

How do I know if AI is citing me?

Ask ChatGPT in search mode, Perplexity and Google's AI answers the exact questions your customer would ask, and see whether you are named or linked. Do it monthly and note which prompts surface you. A paid tracker like Otterly.ai, which has a free trial, will watch it over time, but the manual check costs nothing.

Should I publish on Substack to get found by AI?

For getting cited, it does little. One analysis of 1.7 million prompts found Substack named in 0.07% of AI answers against Medium's 0.36%, because its paywall blocks the crawlers that feed the engines. Substack is strong for building an audience, so keep it, but publish your thinking somewhere crawlable too if you want AI to quote it.