Why does getting cited by AI matter?
AI answer engines now sit between your buyer and your website. Someone asks ChatGPT for the best web development agency, or Google serves an AI Overview instead of ten blue links, and the engine names a small handful of sources. Miss that set and you are invisible for that query, no matter how well you rank the old-fashioned way. In one large analysis of 863,000 keywords and around 4 million AI Overview URLs, only 38% of cited pages also ranked in Google's top 10 for the same query, down from about 76% seven months earlier. A good Google position guarantees you nothing here. It is a different game with different rules.
What actually gets a page cited?
Getting cited is two separate jobs, and a page can nail one and fluff the other. The first gate is selection: does the engine retrieve and list your page at all? That runs on crawl access, indexability, freshness and how strongly the engine recognises you as an entity. The second is absorption: does your wording actually shape the answer? That runs on structure the engine can lift cleanly. Work out which gate a weak page is failing before you go reaching for a fix, because the fixes are not interchangeable.
There is one more thing worth understanding before the list: the engines do not all cite the same sources. They rank and trust differently, and the overlap between them is smaller than most people expect (State of AI Citations 2026, a synthesis of more than 680 million tracked citations, has the per-engine breakdown). So the ten tactics below are the shared foundation, and where you lean hardest depends on which engine your buyers actually use.
The gap is not small. An analysis of 680 million citations found only 11% of domains were cited by both ChatGPT and Perplexity, so being the answer in one is close to no guarantee in the other.
The 10 ways
1. Let the AI crawlers in
Check this first, because if your robots.txt blocks the AI crawlers the selection gate is shut and nothing else on this list gets a look in. Allow GPTBot and ChatGPT-User for OpenAI, ClaudeBot and the anthropic-ai lineage for Anthropic, PerplexityBot, Google-Extended, Bingbot, CCBot and Applebot-Extended. One stray Disallow on any of these is a hard blocker. It is the cheapest thing on the list to fix and the most expensive to leave broken.
2. Put the answer at the top, and make it stand alone
Open every page with a tight answer before any wind-up, because that opening block is what an AI Overview or a voice result reaches for. It has to survive being pulled out of the page with nothing around it, which means naming things explicitly. "We did this for them" is useless to an engine that does not know who "them" is. Lead with the conclusion and let the explanation follow, if it is even needed. A buried answer might as well not be there.
3. Use real tables for anything you compare
Tables get cited far more often than the same information written as prose. Estimates of the size of the effect vary by study, from roughly 2.5x up to 4.2x, but they all point the same way: an engine can read discrete values out of a table and lift them with confidence, where prose makes it guess. The catch is that it has to be a genuine semantic table, not a grid of divs dressed up to look like one with CSS, and never a table saved as an image (those are invisible to crawlers). If a page compares two or more things, the table is not a nice-to-have.
4. Phrase headings as real questions where it fits, and answer them directly underneath
A heading like "Our process" is close to invisible to the way engines scan for answers. Retitle that section as the question a buyer would actually ask, put a direct answer in the first sentence underneath, and suddenly it is extractable. There is a balance to strike, mind you. A page where every heading is a question reads like a robot filled it in, and the human skimming the page matters every bit as much as the crawler. So use the question format where a buyer genuinely asks that question, keep the answer tight, and let the rest read like a person wrote it. You are writing for someone and something at once, not the something on its own. Either way, use one real H1 and a proper H2/H3 hierarchy with actual heading tags, not bold text doing a heading impression.
5. Write so any single paragraph stands on its own
Assume an engine will lift one paragraph and show it with no context around it, then write every paragraph to survive that. Name things. Use real places and real numbers. "The project grew sessions significantly" tells a reader nothing and an engine less. "The rebuild grew organic sessions from 12,000 to 38,000 a month in a year" does the work, because anyone reading it out of context still knows exactly what happened and by how much. That is the difference between a passage that gets cited and one that gets skimmed past.
6. Publish something only you could have written
First-hand, non-commodity content is the single biggest lever there is, and Google and the independent research agree on it for once. A page built entirely from general principle reads like the thousand other pages built entirely from general principle, and the engine treats it that way. Put something real on the page: a named client, an actual number, a project outcome you can point to. A specific, verifiable result ("cut the average page load from 4.1s to 0.9s", "grew organic sessions 3x in eight months") is the kind of thing an engine lifts and a buyer trusts, where a vague "great results" is neither citable nor convincing. Never invent a figure to fill the gap. A made-up stat is worse than no stat, and it has a way of being found out.
7. Build depth across a topic, not one hero page
Google AI Overviews split a query into smaller sub-queries (this is called query fan-out) and cite pages that perform across a whole topic, not just one strong article. It is part of why the top-10 overlap fell to 38%: the engine is pulling from pages that answer the sub-questions, wherever they rank. A single post is one node in a much bigger picture. Link it to the related pages and posts covering the same ground, so the subject has real depth across the site rather than one lonely article carrying the lot. Breadth beats a single hero page, even a very good one.
8. Show a last-updated date, and mean it
Freshness moves AI citation hard. One cross-platform analysis found pages updated within the last 30 days earned roughly 3.2x more AI citations than older ones. The snag is that freshness only counts if the reader and the crawler can see it, so an unstyled last-updated date on the page is the whole mechanism. You cannot claim credit for freshness nobody can see.
9. Get the rest of the web to vouch for you
Citation is only half an on-page job. Engines cite entities they recognise, and that recognition is built from the same facts about you turning up consistently across your own site and the independent sources the models trust. For a web or marketing agency the platforms that matter are the industry ones: Clutch, GoodFirms, DesignRush, Google Business Profile and YouTube, not software directories like G2 or Capterra that index products rather than service firms. YouTube is the quiet standout, where an analysis of 75,000 brands found mentions in video titles, descriptions and transcripts to be the single strongest correlating factor with AI Overview visibility. Keep the facts about you identical everywhere, because inconsistency is what breaks entity recognition.
10. Finish with an FAQ that actually answers things
Close each page with an FAQ covering the related questions the page brushes against but does not build itself around. Each question is a heading, each answer two to four tight sentences that stand on their own. This hands engines more discrete answers to cite and picks up People Also Ask spots in Google along the way. Adding FAQPage schema on top helps Google and Bing understand and present the content, so it is worth doing for the traditional rich-result and extraction benefits. Validate the JSON-LD before it goes live, because malformed schema is worse than none.
