How to Write a Brand Facts Page That AI Models Actually Quote (2026 Guide)

Key takeaways

• A brand facts page works when every fact is a short, self-contained sentence that an AI model can lift without extra context.
• Cover the questions people actually ask about a brand: what it is, who it is for, pricing, features, and how it compares.
• Add Organization schema, sameAs links, and an honest FAQ so machines can verify the page against other sources.
• Landing pages and docs stay cited in ChatGPT for months, so wrong prices or specs will keep getting repeated until you fix them.
• Check results by tracking which claims AI makes about you and which of your pages get cited.

What a brand facts page is (and why it exists)

A brand facts page is a plain page on your own domain that states the basics about your company in a form machines can extract. Name, founding year, headquarters, what you sell, who it is for, pricing, integrations, certifications, leadership, and a few honest comparisons. Think of it as the press kit you would write for a very literal journalist who never calls back to ask a follow-up.

Why bother? Because AI models fill gaps. If your site never says plainly what you cost or who you serve, the model will assemble an answer from review sites, old blog posts, and Reddit threads. Sometimes that answer is fine. Sometimes it says you lack a feature you shipped two years ago.

The data backs up the focus on branded questions. In Promptwatch Data's analysis of ChatGPT citations by content type (Sep 28 to Oct 4, 2026), product pages earned 22.4% of all classified citations and landing pages 16.8%. For branded prompts like "is HubSpot good for startups", product pages (23.3%) and landing pages (22.8%) together earned almost half. Reviews took 5.6%, more than three times their share across all prompts. FAQ pages took 4.8%.

So for brand questions, AI leans on your own core pages first and third-party reviews second. A facts page is a chance to make the first group clean and quotable.

Why AI models quote some pages and skip others

Most content fails to get quoted because it is unextractable, not because it is wrong. A model answering a question is pattern-matching against a query and looking for a self-contained passage it can lift, attribute, and trust. As Firon Marketing puts it in its guide on writing content AI models quote, content written for human flow, with the answer buried three paragraphs into a section, often loses to a thinner but more extractable competitor.

Three habits show up across the sources I read:

• Answer first. The first sentence under a heading should answer the heading.
• One idea per sentence. A sentence with three claims confuses extraction.
• Facts before opinions. Search Engine Land's answer-first guide says models prefer statements backed by numbers or authoritative attribution, followed by context.

A neutral tone matters too. Pages that read like a sales pitch tend to get skipped, while pages that read like a knowledgeable person explaining something get used. A facts page is a good place to practice this, because there is nothing to sell. You are just stating what is true.

What to put on a brand facts page

Start with the questions someone would ask an AI assistant about you. Then answer each one in a sentence or two.

The core identity block

Open with a short definition: what the company is, in one sentence, in plain words. "Acme is a project management tool for construction teams with 10 to 200 employees." No tagline. No "leading provider of."

Then a compact fact list:

• Legal name and trading name
• Founded year and founders
• Headquarters and other offices
• Number of customers or employees, with the date you measured it
• Funding or ownership, if public
• Certifications and compliance (ISO/IEC 27001, SOC 2, GDPR)

Date every number. "1,200 customers as of October 2026" beats "over a thousand happy customers" because a model can attribute it and a reader can judge how stale it is.

Products, pricing, and who it is for

Promptwatch's own recommendation for branded prompts is to make the homepage, product pages, and docs answer pricing, features, and who the product is for. Your facts page should state the same things in one place. List plan names, monthly prices, and what each tier includes. If pricing is custom, say so and state the starting range or what drives the cost.

Include a "best for" and "not a fit for" line. It feels odd to write the second one. It is also the line models tend to trust, because it reads like an honest description rather than ad copy.

Comparisons and category placement

Say what category you are in and name the alternatives people compare you with. The same Promptwatch data shows comparisons earning 5.9% of ChatGPT citations overall and 7.9% on commercial prompts. If you do not state how you differ from competitors, someone else will, and their version gets quoted.

Keep it factual: a short table of three or four differences you can prove, not a claim that you are the best.

A plain-language FAQ

Add 8 to 15 questions a buyer would type. Phrase each heading as the question itself, and answer in the first sentence. "Does Acme integrate with Procore? Yes. Acme syncs projects, budgets, and documents with Procore through a native integration." Then one or two lines of detail.

How to write each fact so it can be lifted

This is where most facts pages go wrong. The facts are present but written like marketing paragraphs.

Compare these:

A few rules I use when editing:

1. Subject first. Start the sentence with the brand name, not "our" or "we". A lifted sentence has to make sense when it appears in someone else's answer.
2. One claim per sentence.
3. Numbers with units and dates.
4. No adjectives you cannot prove.
5. Same wording everywhere. If the facts page says "Team plan" and the pricing page says "Starter", a model has to guess.

Heading structure

Use H2s for sections and phrase them as questions or clear labels, with a short factual answer directly beneath. A heading like "How much does Acme cost?" gives a retrieval system something to match. "Pricing overview" is weaker, though still better than "Our story".

Lists and tables

Use a table for pricing and plan limits, a bulleted list for certifications and integrations. These formats are easy for machines to parse and easy for humans to scan, which is a rare case where both audiences want the same thing.

Add machine-readable signals

The visible text does most of the work, but structured data helps a model confirm it is looking at the right entity.

Organization schema

Add JSON-LD Organization markup to the page with the same facts as the visible text:

{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Acme Inc.",
  "url": "https://acme.example",
  "foundingDate": "2019",
  "founder": { "@type": "Person", "name": "Jane Doe" },
  "address": {
    "@type": "PostalAddress",
    "addressLocality": "Amsterdam",
    "addressCountry": "NL"
  },
  "sameAs": [
    "https://www.linkedin.com/company/acme",
    "https://www.crunchbase.com/organization/acme",
    "https://www.wikidata.org/wiki/Q000000"
  ]
}

The sameAs list ties your site to profiles on other sources. Only include profiles you control and keep current. Mismatched details between your schema and those profiles create exactly the contradictions you are trying to remove.

Keep the facts consistent off-site

I did not find solid research in my sources on how much weight models give schema or entity databases, so treat markup as a consistency aid rather than a guaranteed lever. What is clearer: models compare sources. If LinkedIn says 50 employees, Crunchbase says 200, and your facts page says 120, you have handed the model a reason to hedge or pick the wrong number. Audit your major profiles and review sites quarterly.

Make sure crawlers can read it

A facts page nobody can crawl is a diary. Check robots.txt for AI crawlers such as GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot, and make sure the page renders without JavaScript. Free tools like Promptwatch's Robots.txt generator for AI and AI crawler URL inspector help you check access in a few minutes. An llms.txt file is optional; it is cheap to add, but Promptwatch's markdown data shows markdown files are just 0.05% of AI search citations, so do not count on it for AI search.

Keep it fresh

Stale facts are the biggest risk, and the research makes this concrete. In Promptwatch's content half-life study, which tracked every page ChatGPT cited on July 6, 2026 through Oct 4, 2026, landing pages and docs stayed cited longest. After 30 days ChatGPT still cited 78% of landing pages and 76% of docs. After 60 days it was 69% and 67%. News pages lost half their citations within 36 days.

The practical read: pages like a facts page, pricing page, and docs get cited for months. That is good when the facts are right and costly when they are not. A price you changed in September can still be quoted in December.

A simple maintenance routine:

• Put a visible "Last updated" date at the top.
• Review it monthly, and immediately after any pricing, plan, or leadership change.
• Keep a changelog section for big shifts, such as a renamed plan.
• Update the schema in the same commit as the text.

How to check whether it works

Publishing is half the job. The other half is seeing what AI actually says about you afterwards.

Do it by hand first

Run 15 to 20 branded prompts in ChatGPT, Gemini, Perplexity, and Google AI Mode: "What does Acme cost?", "Who is Acme for?", "Acme vs [competitor]", "Does Acme have [feature]?" Write down every claim that is wrong or missing and which sources the answer cites. Fix the pages that feed the wrong claims, not just the facts page.

Then track it over time

Manual checks do not scale and answers vary by country and model. A visibility platform helps here. Promptwatch pulls the individual claims AI makes about you into Brand Narrative, lets you mark claims as accurate or inaccurate with Fact Check on Business plans and above, and shows which of your crawled pages state the wrong version. Its AI crawler logs show whether bots read the page at all, and Page Tracker shows whether it gets cited.

Other tools in this space cover parts of the job. The directory at bestgeosoftware.com lists many of them.

A short checklist before you publish

1. Opening definition is one sentence with no tagline.
2. Every number has a unit and a date.
3. Pricing and plan names match the pricing page exactly.
4. Headings are questions or clear labels, answered in the first sentence.
5. FAQ covers pricing, fit, integrations, security, and comparisons.
6. Organization schema matches the visible text.
7. Off-site profiles match the page.
8. AI crawlers are allowed and the page renders without JavaScript.
9. A named owner reviews it monthly.

Common mistakes

The most frequent one is writing the page like an About Us story. Origin tales are nice for humans, but a model looking for your headquarters will not mine paragraph six for it.

Second, hiding it. A facts page buried in the footer with no internal links is rarely crawled well. Link it from the homepage, the pricing page, and the About page.

Third, claiming things you cannot back. "The leading platform" invites a model to find a source that says otherwise. State the measurable thing instead, like customer count, rating, or certification, and cite where it comes from.

Fourth, forgetting reviews. For branded prompts, reviews earned 5.6% of ChatGPT citations in the Promptwatch data, well above their overall share. Your facts page cannot control what G2 or Trustpilot says, but you can make sure the facts there are accurate and your profile is complete.

Need help running this at scale?

For a single brand, one person and a spreadsheet can maintain this. For multi-brand or multi-market teams, it turns into an ongoing program of audits, content fixes, and monitoring. Agencies such as 1001 SEO Media handle that work as part of GEO and technical SEO engagements, but the page itself is something most in-house teams can ship in a week.

Write it plainly, date every number, and check what the models say next month. If a wrong claim persists, the cause is usually a different page than the one you just wrote.