A brand fact block for LLMs is a short, dated list of verified facts about your company, placed in plain HTML on your homepage and key pages and backed by matching schema, so AI engines can lift accurate details instead of guessing. It does not force a quote. It removes the excuses for a wrong one.
Key takeaways
- Put 8 to 12 facts in visible HTML, one claim per sentence, each with a number, a date or a named source.
- Mirror the block in JSON-LD, but treat schema as a consistency check, not a lever. Google says no special schema is needed for its AI features.
- Repeat the same wording on the homepage, the About page and each features page, then match it on third-party profiles.
- Check that the bots you want can fetch the page before you judge the block.
- Test with a fixed prompt set, repeated runs and a 14-day gap. One answer proves nothing.
What is a brand fact block, and why do engines need one?
Engines build answers from pages they retrieve and from what their models learned earlier. When your own page states a fact in a vague or scattered way, the engine fills the gap from reviews, old press or a similarly named company. A fact block gives the retrieval step one clean passage to take.
There is evidence that quotable material matters. The GEO paper by Aggarwal et al. found that adding citations, quotations from relevant sources and statistics to content raised its visibility in generative engine responses, by up to 40% on its benchmark. That was a research benchmark, not your category, so treat 40% as a ceiling, not a forecast.
Yext reports a closer analogue. In its own A/B tests on Location.com between October 2025 and January 2026, pages carrying its signed Brand Certified Facts markup saw a 37% click lift on Yahoo, a 35% lift on Bing and a 9.2% lift in Google Gemini citations, against pages without it (Yext, 2026). It is a vendor test on directory pages, with a proprietary method, so it shows direction rather than what a SaaS homepage will get.
What none of this proves is that a block on its own makes ChatGPT quote you. Treat it as one input and measure it, as described below. For how engines choose between pages, see how AI engines choose citations.
What should a brand facts page include for AI engines?
Include only facts that a buyer asks about and that you can prove on the same page or one click away. Each fact needs an owner inside the company, because someone must update it.
| Fact | Example wording pattern | Why an engine needs it | Who owns it |
|---|---|---|---|
| Full name and what it is | "[Brand] is a [category] for [audience]." | Defines the entity in one quotable sentence | Brand or product marketing |
| Disambiguation | "Not affiliated with [similarly named company]." | Stops mix-ups with namesakes | Brand |
| Founded or launched | "Launched in [Month Year]." | Dates are easy to get wrong | Brand |
| Who it serves | "Built for [role] at [company type]." | Drives "best for" answers | Product marketing |
| Core capabilities | Three to five verbs, each with a count | Gives exact figures to quote | Product |
| Pricing and trial | Plan names, prices and the pricing URL | A wrong price misleads buyers and is quick to correct at the source | Finance or growth |
| Integrations and limits | Named tools, plan limits | Answers fit questions | Product |
| Proof and sources | Link to docs, a published study or a certificate | Lets the engine cite you, not paraphrase a reviewer | Content |
| Last reviewed date | "Facts checked on [date]." | Signals freshness | Whoever runs the review |
What to leave out
Leave out promotional adjectives, unsourced rankings and customer counts you cannot evidence. A sentence such as "supports 13 AI engines" can be checked and quoted; a vague boast about power or quality cannot. Also leave out anything that changes weekly unless the block links to the live page that holds it.
How do I write the block so engines quote it accurately?
- Write each fact as a complete sentence with the brand named in full. A passage lifted out of context must still make sense. "It launched in May 2026" fails; "Noma by Nurdd launched in May 2026" works.
- Keep one claim per sentence. Compound sentences get half-quoted, and the half that survives may be wrong.
- Attach the number to its unit and date. "Starter is $79 per month, in USD, as checked on 10 November 2026" cannot be misread.
- Use a table or list, not a graphic. Text in an image is invisible to most crawlers.
- Link each claim that has a source. A visible link to documentation or a study gives an engine something to cite.
- Add a review date and a name. "Reviewed on 10 November 2026 by [role]" shows the block is maintained.
The tone is closer to a reference entry than to marketing copy. A useful test: could a journalist paste the sentence into an article without editing? If not, rewrite it.
A worked example
Here is a block built only from facts Noma by Nurdd publishes about itself, as an illustration of the pattern:
Noma by Nurdd is an answer engine optimisation (AEO) and AI visibility platform. It asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, the questions a brand's buyers ask, every day. Noma by Nurdd launched in May 2026 and is not affiliated with Noma Security. Every plan starts with a 3-day free trial; a card is taken at signup so billing can start on day 3.
Each sentence names the entity, holds one claim and could be quoted alone. The disambiguation line exists because a namesake in security software would otherwise be a likely source of confusion. If your name is shared, see the entity audit for brand confusion.
Which pages should carry the block?
Do not paste the same block everywhere. Use a short version where a page already answers a narrow question, and keep the full version on one canonical page.
| Page | Block size | Facts to prioritise | Link to |
|---|---|---|---|
| Homepage | 4 to 6 facts | What it is, who it serves, disambiguation, launch date | About and pricing |
| About or "Cite us" page | Full, 8 to 12 facts | All of the table above, plus the review date | Docs, press kit, profiles |
| Features page | 3 to 5 facts | Capabilities with counts, integrations, limits | Help article or docs |
| Pricing page | Plan table with dates | Prices, limits, trial terms | Terms and billing help |
| Product or location pages | Per-item facts | Specs, availability | Merchant or listing source |
Place the block above the fold or right under the page's opening paragraph, in the HTML source and not injected by script late. Keep it identical in wording wherever it repeats; contradictory versions are worse than none.
How should I pair the block with schema?
Schema repeats the block in machine-readable form so that the two cannot drift apart. Google's own documentation is direct that it is not required for AI features: its guidance says "You don't need to create new machine readable files, AI text files, or markup" to appear there (Google Search Central, as checked November 2026). So pair them for consistency and for the engines and tools that read JSON-LD, and do not expect markup to compensate for thin text.
| Block fact | Schema.org property | Note |
|---|---|---|
| Full name | Organization name, legalName, alternateName | Use the spellings people search for |
| What it is | description | Match the block's first sentence |
| Launch date | foundingDate | Same date as the visible text |
| Official profiles | sameAs | Only profiles you control or that are verified |
| Contact | contactPoint | Must match the visible contact details |
| Prices | Offer on Product or SoftwareApplication | Only where the price is shown on the page |
| Questions answered | FAQPage | Only for questions visible on the page |
Google's Organization structured data documentation lists the administrative properties it reads, such as address, contact information and business identifiers. The rule from Google's general guidelines still applies: mark up what users can see. Build and validate the JSON-LD with the free Noma schema markup generator, which prefills from your site and checks the result. More detail is in schema markup for AI search.
Can the engines actually read the page?
A block that crawlers cannot fetch does nothing. OpenAI's crawler documentation says that OAI-SearchBot and GPTBot robots.txt tags let site owners manage how their sites work with OpenAI products, and they are separate controls (OpenAI, as checked November 2026). A site can allow OAI-SearchBot for search and decide on GPTBot separately. Check your robots.txt and your CDN rules with the free AI crawler checker before running any test. If your block sits in a tab, an accordion that loads on click or a client-side widget, view the page source and confirm the text is there.
An llms.txt file can carry the same facts, but treat it as an extra copy rather than a substitute. You can check yours with the llms.txt checker guide.
How do I test whether engines pick the block up?
Run a before-and-after test with a fixed prompt set. Answers vary run to run, so a single check misleads. In Noma's own analysis of 2,545 answers from ChatGPT, Gemini, Perplexity and Claude, two in three cited URLs appeared in one run only (Noma by Nurdd, November 2026). Expect noise.
- Write 10 to 15 fact prompts where the answer is in your block: "What does [brand] cost?", "When did [brand] launch?", "Is [brand] the same as [namesake]?", "Which engines does [brand] track?" Add 5 unbranded prompts a buyer would ask. See how to choose prompts to track.
- Record a baseline before publishing. Run each prompt at least three times per engine on separate days. Log whether the answer names you, whether each fact is correct, and which URLs it cites.
- Publish the block and the schema together and note the date. Change nothing else on those pages.
- Wait 14 days, then repeat the same prompts the same number of times.
- Score three things: fact accuracy (correct facts over facts asked), citation of your own URL, and mention rate on the unbranded prompts.
- Treat small moves as noise. Write down your numerator and denominator, and only call a change real if it is well beyond the run-to-run spread you saw in the baseline. The AI visibility measurement standard sets this out.
For a quick manual start, the free AI visibility checker shows whether ChatGPT, Gemini, Claude and Perplexity recommend your brand and who they name instead. The AI brand fact check walks through matching each answer claim against your facts.
Reading the result
| What you see | Likely cause | Next step |
|---|---|---|
| Facts correct, your URL not cited | The engine knows the fact from elsewhere | Link facts to a source page; check which pages it cites |
| Old fact still appears | Stale third-party page or cached model knowledge | Find the source and correct it; see the correction runbook |
| One engine right, another wrong | Different retrieval sources | Check that engine's bot access and cited pages |
| No change after 14 days | Page not recrawled, or block not visible in source | Fetch the page as a bot would and re-test |
Where Noma fits
Noma by Nurdd is an answer engine optimisation and AI visibility platform. It asks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, your buyers' questions every day, and shows whether your brand is named, how it is described, which competitors appear instead and which pages the answers cite. That is the before-and-after test above, run on a schedule.
See it on your own brand: book a demo or start the 3-day free trial. A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged.
Common mistakes
- Publishing facts you cannot prove. An unsourced claim invites the engine to prefer a third party.
- Letting the schema and visible text disagree. Different prices or dates in each make a wrong answer more likely.
- Treating it as a one-off. A block with a date from last year signals neglect. Set a monthly review.
- Judging from one answer. Run repeats, as above.
- Ignoring off-site facts. Your block is one source among several; align your directory and review profiles with it.
How we researched this
We reviewed platform documentation from Google Search Central and OpenAI, the published GEO research paper, a vendor's published A/B test and several current practitioner articles on brand facts and LLMs, noting what they leave out: a concrete block template, a page-by-page placement plan and a repeatable test. Facts and prices were checked on 2026-11-10. Where a number comes from a vendor's own test, we say so.
What to do next
Draft your block this week: list eight facts, assign an owner to each, and publish it with matching JSON-LD. Baseline your prompts before it goes live. Then book a demo to see the results across 13 AI engines, or start the 3-day free trial.
Read next: how to optimise for Meta AI.
See what AI engines say about your brand
Free · no account · or start a 3-day trial
Questions people ask
How do I get ChatGPT to quote accurate facts about my company?
You cannot force a quote, but you can make accurate facts easy to retrieve. Publish a short, dated facts block in plain HTML on your homepage and About page, allow OAI-SearchBot in robots.txt, and keep third-party profiles consistent. Then test with a fixed prompt set over several runs.
What should a brand facts page include for AI engines?
Include the full brand name and what it is, a disambiguation line if your name is shared, launch date, who it serves, core capabilities with counts, prices and trial terms, links to proof, and a review date. Write each as a complete sentence naming the brand.
Does schema markup make AI engines quote my facts?
Not by itself. Google says no special schema.org markup is needed for its AI features. Schema is still worth adding so the visible block and machine-readable data match, which reduces contradictory facts.
Where should the fact block go on my site?
Put a short version of 4 to 6 facts on the homepage and 3 to 5 on each features page, with the full block on one About or cite-us page. Keep it in the page source, near the top, and word it identically wherever it repeats.
How do I test whether engines use my fact block?
Run 10 to 15 fact prompts on each engine at least three times before publishing, then repeat after 14 days. Score fact accuracy, citations of your URL and mention rate, and ignore small moves that fall within the normal run-to-run variation.
How often should I update a brand fact block?
Review it monthly and whenever a price, plan limit or product claim changes. Update the visible review date and the schema together, and correct any old pages that still show the previous figure.
