A buyer with a real budget opens ChatGPT and types a full sentence: which company can solve this problem for a manufacturer like us? The answer names three firms and explains why each one fits. Nobody scrolls a list of ten links. The shortlist is drawn inside the answer, before your sales team knows the search happened.

If your company is not in that answer, your Google ranking does not help you there. The usual advice makes it worse: chop articles into pieces "for the AI", edit robots.txt, buy a tool that promises a position. The real method is less exciting and works in a fixed order: let the engines read you, describe yourself consistently, answer first, publish what nobody else has, get third parties to confirm it, and measure with four numbers read together.

How do you get cited by AI?

You get cited by AI when an engine can read your pages, understand who you are, find a direct answer to the buyer's question and confirm it somewhere else. In practice that means open access for live-answer crawlers, one consistent entity description, answer-first pages, original data, third-party corroboration and measurement by citations, not clicks.

The work is the same for ChatGPT, Copilot, Perplexity, Gemini and Google's AI Overviews. Google's official guide, "Optimizing your website for generative AI features on Google Search", says it plainly: optimizing for AI Overviews and AI Mode is SEO, because those features run on the ranking and quality systems of Search. For the differences between the three disciplines, read AEO vs SEO vs GEO.

Why are buyers asking AI who to hire?

Buyers ask AI who to hire because a generative answer does the comparison for them and hands back names with reasons. The demand shows up in search data. In Google Ads Keyword Planner, US monthly searches for "how to rank in chatgpt" went from 140 in June to 320 in August 2026, and "ai seo agency" went from 1,900 in September 2025 to 2,900 in August 2026. Keyword Planner volumes are fixed brackets, not exact averages.

The questions are changing shape too. Between September 13 and October 10, 2026, Sell with Marketing's Search Console logged queries written by an assistant, not typed by a person: "who are the best manufacturing marketing agencies in the united states?" (average position 1), "recommend a digital marketing agency for a b2b manufacturer." (position 2.3) and "where can i get expert search engine optimization for google and ai search engines?" (position 3.0, 121 impressions). All had zero clicks: a machine read those results, not a person.

What are the steps to get your company cited by AI?

The steps, in order, are open access, readable HTML, one consistent entity, answer-first pages, original evidence, accurate structured data and third-party corroboration. Each step depends on the one before it: a great page that crawlers cannot read earns nothing.

Can AI crawlers actually read your site?

AI crawlers can only cite what they can fetch and parse. Separate two families. Live-answer crawlers (OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-SearchBot, Claude-User) fetch pages to answer a question right now, so blocking them means disappearing from those answers. Training crawlers (GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot) collect data for model training, so blocking them is a business decision, and a legitimate one.

Robots.txt is not the whole story. A firewall can block AI crawlers even when robots.txt allows them, and Cloudflare has blocked AI crawlers by default on new domains since 2025. The reliable check is to request your pages with each real user agent.

Then check rendering. Many AI crawlers read the HTML your server delivers and do not execute JavaScript, so content that appears only after JavaScript runs can be invisible to them. Three checks for today:

  1. Open yourdomain.com/robots.txt and look for a Disallow under any live-answer crawler, or a blanket Disallow under the wildcard user agent.
  2. View the source of your homepage and search for "application/ld+json". If nothing appears, a crawler reading raw HTML sees no structured data about your company.
  3. Disable JavaScript and reload the page. If the main content disappears, that is close to what many AI crawlers see.

Is your company described the same way everywhere?

An engine names a company with confidence when every source agrees on what it is. Use the same name, category, services, markets and locations on your website, LinkedIn page, Google Business Profile, Bing Places, industry directories and executive profiles. Every contradiction (a different tagline here, an outdated service list there) is a point where the model is unsure whether it is still reading about you.

Does every page answer one buyer question first?

A page earns citations when its answer sits in one self-contained block that an engine can lift without rewriting. Phrase headings as the buyer's question. Put a 40 to 60 word answer directly under each one. Make every paragraph understandable on its own, with no "this" or "the above" pointing back: engines extract paragraphs, not pages.

Replace adjectives with facts. "Fully certified" says nothing; "ISO 9001 and AS9100 certified" can be checked. "Fast turnaround" says nothing; your actual standard lead time can be quoted. Specific sentences get extracted. Hedged marketing copy gets skipped.

What do you publish that nobody else has?

Original evidence is the one thing a competitor cannot copy and a model cannot generate: your own measurements, benchmarks, field experience or a named method with a published sample. Attribute each claim to a real person with a title and a verifiable profile.

Sell with Marketing does this with its own numbers. In Microsoft Bing Webmaster Tools, AI Performance report, from February 22 to August 19, 2026, Copilot cited Sell with Marketing 2,007 times. In 33 of 46 buying questions, SWM held 30% or more of the citation share, and four pages carried 90% of the citations. The full breakdown is in what we measured on AI visibility.

Does your structured data describe real facts?

Structured data helps an engine understand what is already true on the page; it does not create visibility by itself. Generic markup a CMS inserts on its own (WebSite, a bare Organization) tells a model little. Markup with real facts does more: Organization with your actual address, markets and profiles, Person for authors with their credentials, Service or Product with real specifications. Never mark up what is not visible on the page, validate every block, and change dateModified only when the content really changes.

Do third parties confirm what you say?

Your own site is necessary but not sufficient. Engines weigh independent sources because they are harder to fake: industry directories, trade press, associations, partner listings, interviews and reviews. The goal is not link volume but several independent places telling the same specific story your site tells. Start with the sources engines already cite for your buyers' questions.

What should you do at each step, and how do you check it?

  • 1. Access — What to do: Allow live-answer crawlers and review firewall and CDN rules; How to check it: Request your pages with each real user agent
  • 2. Rendering — What to do: Serve key content in the HTML, not only after JavaScript; How to check it: Reload the page with JavaScript disabled
  • 3. Entity — What to do: Use one name, category and service list everywhere; How to check it: Compare site, LinkedIn, directories and profiles side by side
  • 4. Answers — What to do: Write question headings with a 40 to 60 word answer below; How to check it: Read each paragraph alone and confirm it still makes sense
  • 5. Evidence — What to do: Publish original data with a named author; How to check it: Ask what the page has that the other ten results do not
  • 6. Structured data — What to do: Mark up real facts that are visible on the page; How to check it: Validate every JSON-LD block
  • 7. Corroboration — What to do: Get listed and quoted where engines already look; How to check it: Note which sources engines cite for your buyers' questions

What does not work?

Three tactics get sold hard and move nothing:

  • Chopping articles into pieces "for AI". Google's guide says there is no need to split content into chunks or create special formats for AI.
  • Robots.txt edits sold as urgent. If your wildcard rule already allows crawlers, naming each AI crawler is housekeeping, not a fix. The same goes for llms.txt: optional, never step one.
  • Promised positions or guaranteed citations. Generative engines do not give the same answer twice, so every result is a snapshot with a date. No engine delivers a fixed spot in ChatGPT.

A fourth: calling yourself "the leader". A self-declared superlative gives an engine nothing to verify. Replace it with the fact behind it, or delete it.

How do you measure whether AI cites you?

You measure AI citations with four numbers read against each other, never with referral sessions:

  1. Citations by engine and time window. Bing Webmaster Tools has an AI Performance report with Copilot citations, the only quantitative citation source available for free.
  2. Share of the answer on buying questions. Not how often you are cited, but how much of the answer you occupy when someone asks who to hire.
  3. The branded search curve in Google Search Console.
  4. Forms, WhatsApp and CRM records, matched by date and time against the other three.

Expect messy data. For SWM, the three reports in Bing's tool did not match (2,007, 1,894 and 704) because they aggregate over different windows. Those citations sent almost no visits: the effect showed up later, as branded searches and "direct" traffic. That is a correlation in one company, not a proven cause. Judge the channel by AI referral sessions and you will conclude it does nothing.

Google is harder to read. The Search Console API does not separate AI Overviews or AI Mode as a search type: SWM tested every variant against its own property and all returned errors. No reliable automated dashboard for Google's AI features exists today, so a fixed set of buyer prompts, run on dated occasions across several engines, fills the gap. To see where you stand today, start with does AI recommend your company? and the AI visibility sprint.

Frequently asked questions

How long does it take to get cited by AI? There is no fixed timeline, and anyone who gives you one is guessing. Access and rendering fixes take effect when crawlers next fetch your pages. Entity consistency, original data and third-party mentions build over months. Track the trend across dated runs of the same prompts instead of judging a single answer.

Is getting cited by ChatGPT different from ranking in Google? Partly. Google's AI Overviews and AI Mode run on the ranking and quality systems of Search, so good SEO feeds them directly. ChatGPT and other assistants also lean on live crawls and third-party sources. The groundwork is shared: readable pages, a clear entity, direct answers and corroboration. Mostly the measurement differs.

Should you block AI crawlers? Block training crawlers only if your business decides it does not want its content used for model training. Never block live-answer crawlers such as OAI-SearchBot, PerplexityBot or Claude-SearchBot by accident, because that removes you from their answers. Check your firewall as well as robots.txt, since either one can block them.

Can you see AI visibility in Google Search Console? Not as a separate report, because the Search Console API does not separate AI Overviews or AI Mode as a search type. You can watch your branded search curve there, and Bing Webmaster Tools reports Copilot citations. Combine both with your CRM records, read by date and time.

Getting cited by AI is not a trick: it is the evidence a buyer's assistant needs to name you, built in order and measured honestly. To find out which step is your weakest, request a diagnostic.