How to get cited by AI

Most advice about "getting cited by AI" is guesswork dressed up as strategy. This isn't. I run three websites, a gaming data site, a college financial aid site, and a family storytelling site. Over the past quarter, AI engines cited those sites more than 15,000 times across 248 different pages. I pulled every citation report, tested my own pages against ChatGPT, Claude, Gemini, and Perplexity, and compared the pages that win constantly against the hundreds that never get picked once.

This page is the formula that came out of that data. It's written for the site owner starting at zero, which is most of you. And I'm following my own playbook as I write it, so you can see the formula working on the page that teaches it.

01Check your scoreboard before you change anything

Here's something almost nobody does. Bing Webmaster Tools has a free report called AI Performance that shows you, page by page, every time Microsoft's AI surfaces cited your site. Most site owners have never opened it. Some of you are already getting cited and have no idea. Others will confirm the zero, and now you have a baseline instead of a vibe.

Do this today:

• Verify your site in Bing Webmaster Tools (free, takes minutes)
• Open the AI Performance report and export all three views: per-page, daily, and the queries that triggered citations
• Write down your number, even if it's zero

One of my sites sat at zero for 64 days after reporting started. Another got cited on day one. Both were fixable situations, but only because I was looking at the actual scoreboard instead of guessing.

02Pick topics the AI can't answer from memory

This is the gate, and it's the step everyone skips. When someone asks ChatGPT a question, the model first decides whether to search the web at all. If it thinks it already knows the answer, it never searches, and if it never searches, nobody gets cited. Not you, not your competitor, nobody.

I tested this directly. I ran the same questions through ChatGPT, Claude, Gemini, and Perplexity with web search turned on and inspected the raw responses. On evergreen how-to questions, three of the four engines answered straight from memory, zero sources, over and over. On questions about things that change, current data, recent events, niche numbers, the engines searched almost every single time. Perplexity was the one exception, it searches on basically everything.

Your topic forces the AI to search if at least one of these is true:

• It changes on a cycle. Prices, rates, rankings, seasonal data, anything with a version number. My biggest page is a rankings page that shifts every few weeks, and it has 6,000+ citations to show for it.
• It's newer than the model's training. A shutdown, a launch, a policy change. A single "what happened to [company that shut down]" page became the most-cited page on one of my sites, beating pages I'd spent months on.
• It's specific data nobody memorizes. Per-product specs, per-school numbers, per-city rates. I have 100+ different data pages cited off one template.
• The honest answer needs today's numbers. If a truthful answer requires a current figure, the model has to go look.

The brutal counter-example from my own data: I have beautifully built evergreen advice pages, 2,400+ words, FAQs, answer blocks, real sources. ChatGPT, Claude, and Gemini cited them exactly zero times, because they never searched the topic. Great pages, wrong game.

For a small business, this means: your "10 tips" blog post is probably invisible to AI, but your current pricing comparison, your local rate sheet, your "is X still open / what replaced X" page, those force a search. Publish what the model can't already know.

03Win regular search, because that's what the AI reads

Here's the unfashionable truth: when an AI does search, it mostly cites whatever its search backend returns near the top. My most-cited page ranks around position 2 on Google for its main query. That rank, not any AI trick, is why the engines keep grabbing it.

So the boring 80% of "AI optimization" is just SEO that works:

• Rank for the question cluster you want to be cited on. If you're on page 4 of search, no formatting fixes that.
• Make sure your content exists without JavaScript. Bing's crawler renders JavaScript, but most AI fetchers don't. Quick test: view your page's source and search for a fact only your real content contains. If it's not there, some engines literally cannot read you.
• One clean domain, boring predictable URLs. AI engines guess URLs. I keep redirects for the URL forms they guess wrong, and those redirects catch real citation traffic.
• Don't assume you're too small. In my live tests, tiny single-topic sites beat big brands for citation slots, repeatedly. The bar is relevance and rank, not fame. This is genuinely good news for small businesses.

04Make your answer liftable in one grab

Once demand exists and the engine can find you, structure decides what gets quoted and protects you across every engine. On every one of my winning pages, the same pattern shows up:

• An answer-first block. Two to five sentences directly under your headline that fully answer the question, in plain language, naming names. If an AI quoted only that block, the reader would be satisfied. (Scroll up. This page opens with one.)
• One structured payload. A table, a ranked list, or a key-facts box. Give the machine a single clean unit to lift instead of making it assemble your answer from twelve paragraphs.
• A visible date. On the page, where a human and a machine can both see it, and matching the date in your metadata. AI engines are biased toward fresh, and an undated page reads as stale.
• An FAQ that actually exists on the page. Real visible questions and answers, with schema markup that mirrors them. Never schema-only FAQs that don't appear on the page.

One honest note, because my data forced it on me: structure alone never made a page win. My biggest page earned most of its 6,000+ citations before I added any of this to it. Structure is the price of admission and the quote-shaper. Steps 2 and 3 are what win.

And on length: stop padding. In my controlled 751-page test, word count predicted nothing. Adding structured, liftable units had a real effect. Adding words did not. Write the complete answer, structure it hard, and stop.

05Refresh on a schedule, not on inspiration

The pages that get cited constantly are the ones the engines learn to trust for current answers. My biggest winner updates weekly from a single data source, and its update date changes automatically because the update actually happened. That cadence is the moat, competitors would have to maintain the same data pipeline to compete.

• Own one data asset in your niche that must be maintained to stay true, prices, availability, rankings, rates
• Put the refresh on a calendar, weekly or monthly, and update the visible date only when the content really changed
• Ping the search engines when you publish (IndexNow is free and takes an afternoon to set up). My site with publish pings got cited on day one of reporting. My site without them took 64 days to its first citation.

06Catch demand windows

The single most-cited page on one of my sites answers "what happened to [platform that shut down]." I published it within weeks of the shutdown, with a clear timeline and honest "here's what we know" framing, while the official communication was thin. It now out-earns pages I spent ten times longer building.

When something shifts in your niche, a competitor closes, a price changes, a rule updates, a product launches, ship the definitive factual page that week. That's exactly when the AI's memory is guaranteed to be wrong and the search gate is wide open. Expect a harvest, not an annuity, that demand decays. Take it anyway.

07Measure it like money

• Monthly: export the Bing AI Performance reports and track your cited pages and totals
• Quarterly: run your ten most important customer questions through ChatGPT, Perplexity, and Gemini, three times each. Log who got cited, and log whether the AI even searched. Your topics' no-search rate is the single most strategic number you can know.
• In your analytics: track referrals from chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, and copilot.com. One of those sources is easy to miss and was almost half of one site's AI traffic.
• Set expectations: citations run far ahead of clicks. A lot of the value is your name in the answer, brand exposure at the moment of the question. The clicks I do get concentrate on interactive tools and calculators, not articles. If you want AI visitors, give them something to come do.

08What does NOT get you cited (I tested it, so you don't have to)

This is where I save you real money, because entire agencies are selling this list right now:

• JSON-LD schema as a citation driver. My most-cited page earned thousands of citations with zero schema. Across my sites, my uncited pages actually carry more schema than my cited ones. Ship schema for regular SEO hygiene, don't pay anyone who promises it wins AI citations.
• Server-side rendering as a requirement. Static, server-rendered, and pure client-side JavaScript pages all got cited in my data. What matters is whether each engine's crawler can see your facts, not your tech stack.
• Word count. Statistically meaningless in my controlled test. My 84% stat again: the vast majority of my deepest pages earned nothing, because depth without demand is a book nobody asked for.
• Question-shaped headings. Zero percent of my top-cited templates use them.
• llms.txt. All three of my sites have polished ones, including pages with zero citations. It explained nothing.
• Splitting one strong page into many small ones. I tested this prospectively. The combined hub pages earned roughly 17 times more citations per page than the fragments. When people ask broad questions, build one dense page, not twenty thin ones.

Every item on that list is something the industry charges for. In 15,000 citations of my own data, none of it selected the winners. Topic selection and search presence did.

09Your first 30 days, if you're starting from zero

• Week 1Verify in Bing Webmaster Tools, pull the AI Performance report, get your baseline. Set up IndexNow.
• Week 1Run your ten most important customer questions through ChatGPT and Perplexity. Note who gets cited and whether the AI searched at all.
• Week 2Pick one topic from Step 2 that forces a search, current data, a recent change, or specific numbers only you maintain. Confirm people actually search it.
• Week 3Build one page: answer-first block, one table or key-facts box, visible date, real FAQ. Complete, not long.
• Week 4Publish, ping, and put the refresh on your calendar. On an established site, my new pages started earning citations in 6 to 8 days. Then repeat with page two.

10FAQ

Pick one topic that forces a search, ship one complete page, and check the scoreboard a week later. If you want help applying this to a site you own, that's the work I do.

Last updated: July 2, 2026