How to Get Cited by ChatGPT and Perplexity

Extractability is a writing discipline. Open every important page with a direct 40 to 60 word answer to the question it targets, before any elaboration, because that is the block an assistant is most likely to quote. Phrase your headings as the questions real buyers ask, then answer each one plainly and completely underneath. Use short declarative sentences, clear definitions, and, where it helps, lists and comparison tables that a model can parse without guesswork. Structured data such as FAQPage and Article markup can state what the page contains in machine-readable form, but Google says no special schema is needed to appear in its AI features, so treat it as housekeeping, not a lever. Support claims with specific, sourced facts rather than vague adjectives, since concrete statements are more quotable and more credible. Avoid burying the answer inside long narrative or hiding it behind a click. The test we use is simple. Read the first two sentences under a heading out loud. If they answer the question on their own, an assistant can quote them. If they set up the answer but make the reader keep going, they are not extractable yet, and we rewrite until they stand alone. A page that passes this test does triple duty. The person skimming finds the answer. The search engine has a clean snippet to pull. The assistant has a self-contained block to quote and attribute. Content built like this can be pulled as a snippet, a People Also Ask answer, or an assistant citation, all from one honest structure. You are not writing twice, once for people and once for machines. You write once, clearly, so both find the answer in the same place.

An assistant reads far more than your website when it decides whether to name you. It triangulates across everything it can find, and if those sources disagree or barely mention you, it stays quiet. Building your entity means making your business a settled, consistent fact on the open web. Keep your name, address, and phone identical across every directory and listing. Earn genuine citations and credible mentions, appear in the reputable listicles and roundups for your market, and pursue real press rather than manufactured noise. Encourage authentic reviews and answer questions in the communities where your customers already talk. None of this is a trick, and coordinated fake activity backfires, since platforms and models increasingly detect it. When we rebuild a contractor's off-site presence, we start with consistency, not volume: a plain list of every place the business is already mentioned, then reconcile the name, address, and phone one row at a time until they agree. We are not chasing a hundred new links. We are making sure the listings that already exist all agree on who the business is, where it works, and how to reach it. That agreement is the consistency a model can verify. The durable path is a real, consistent, well-described presence in the places AI reads. Do that, and the model has a trustworthy record to draw on when someone asks who to call. Try to fake it, and you are building on a foundation the platforms are getting better at knocking down every month.

You cannot be cited by a system that cannot read you. Two technical mistakes quietly remove businesses from AI answers. The first is blocking the crawlers that feed AI answers, whether deliberately or through a default rule left over from a staging site. Each AI company runs more than one crawler, and they do different jobs. OpenAI documents that OAI-SearchBot is the crawler that surfaces sites in ChatGPT search, while GPTBot collects content that may be used to train its models, and robots.txt controls the two separately. Perplexity documents PerplexityBot as the crawler that surfaces and links sites in its search results. Google’s AI Overviews and AI Mode draw on the ordinary Search index, so Googlebot access is what counts there; Google-Extended governs use of content for Gemini training and grounding in some other Google systems, not inclusion in Search. You can decide on training separately, but never block the search crawlers. Google also states that no special AI text file or schema markup is required to appear in its AI features, so treat an llms.txt file as optional housekeeping, not a lever. The second mistake is hiding content behind JavaScript. If your key answers only appear after a script runs, a crawler that does not execute that script sees an empty page. Put your important content in server-rendered HTML so it is present the instant the page is fetched. Both problems usually happen by accident. A developer ships a template with a blanket crawler block left on from a staging site. A modern framework renders the whole page in the browser, so the answer a human sees never reaches the crawler. Neither is malicious, and both make you invisible to the exact systems you want quoting you. That absence carries a growing cost: Gartner predicted in February 2024 a 25 percent drop in traditional search engine volume by 2026 as AI chatbots and virtual agents absorb queries, so a business a crawler cannot read loses the citation and the click at the same time. Blocking crawlers to protect content is a false economy: it does not stop scraping so much as keep you out of the answer. Openness plus clean, server-rendered HTML is the price of being quoted.

You manage what you measure, and AI visibility needs its own scorecard. Traditional rank tracking tells you where a page sits in a list of links, but it says nothing about whether an assistant names you. Track three new numbers instead. Mention rate, how often each engine references your business for the prompts your customers actually use. Citation rate, how often it links your specific page as a source. And share of voice, how you compare with competitors inside those generated answers. You can monitor this with dedicated AI-visibility tools or, at minimum, by running the same buyer prompts across ChatGPT, Perplexity, and Gemini on a regular cadence and logging who gets named. The way we do it is unglamorous and repeatable. We write down the ten questions a real customer would ask before hiring, we ask each engine those questions on a fixed schedule, and we record who was named and who was cited. Over a few months the trend line, not any single answer, tells the story. Two honest caveats come with this. First, no reputable firm can guarantee a specific model output, because the systems are probabilistic and change often. Second, we do not publish per-city numbers we cannot source, so we report the mention and citation rates we can actually measure, plainly, whether they rise or fall. What you can do is build every signal the models weigh, then watch those rates move, and if you want to see where you stand before committing to anything, the audit that maps it is free and yours to keep. A number you track honestly, even when it dips, is worth more than a promise no one can keep.