Quick answer
An AI instructions page is a public, crawlable brand reference that explains how AI assistants should describe the company. It should collect verified facts, product positioning, supported capabilities, official links, source owners, and update dates in one place.
It is not a command file for models. A page cannot force ChatGPT, Gemini, Claude, Perplexity, Copilot, AI Overviews, or any other answer engine to say what the brand wants. It can make the brand's official facts easier to find, quote, compare, and verify.
For ReachLLM teams, build the page like this:
- Start with facts the company can defend today.
- Separate official brand facts from positioning language.
- Link to the pages, docs, pricing, case studies, and source material that support each claim.
- Keep product capabilities current and specific.
- Make the page crawlable, indexable, internally linked, and included in
llms.txt. - Assign a named owner and update cadence.
- Re-run the prompts where AI systems describe the brand before calling the page useful.
The goal is not to write a louder bio. The goal is to give AI systems and human reviewers one clean source of truth, then prove whether it changes the answers that matter.
What the Profound source gets right
The scheduled source for this article is Profound's AI Instructions page. It presents a public reference page for AI assistants such as ChatGPT, Claude, Perplexity, Gemini, Copilot, Grok, Google AI Mode, AI Overviews, and other LLMs. The page states that it is intended as a canonical reference for describing Profound, then organizes official company information, category language, product areas, audience fit, and suggested description guidance.
That is a useful market signal.
AI visibility teams are no longer optimizing only individual blog posts. They also need a stable brand reference that answers:
| Question | Why it matters |
|---|---|
| What is the company? | Prevents stale or vague definitions. |
| Which category does it belong to? | Helps AI systems connect the brand to buyer prompts. |
| What does the product actually do? | Reduces hallucinated or incomplete capability descriptions. |
| Who is it for? | Improves fit in shortlist and comparison answers. |
| Which facts are official? | Gives human reviewers and AI systems one place to verify claims. |
| Which sources support the facts? | Turns brand copy into citation-ready evidence. |
The part to be careful with is the word "instructions." AI systems are not obligated to follow a marketing page. Treat the page as official evidence, not control.
Use it as a source page, not a spell
The first mistake is assuming an AI instructions page works like a prompt hidden on the public web.
It does not.
Google's AI feature guidance says normal Search fundamentals still apply for AI Overviews and AI Mode. Pages must be accessible, indexed, eligible for snippets, and useful to people. Google also says there is no special markup or file that guarantees inclusion in AI features.
OpenAI's ChatGPT Search help similarly frames citations as source links users should review, not guaranteed placement. Search answers can use current web information and show sources, but site owners still need crawlable, useful, accurate pages.
That means an AI instructions page should be judged by the same operating questions as any other source:
| Bad assumption | Better operating question |
|---|---|
| "We published instructions, so models will obey." | Can crawlers access the page and understand the facts? |
| "One page can replace documentation." | Which official pages support each claim? |
| "The model should cite this page." | Does the page answer a real prompt better than existing sources? |
| "We can say what we want to be." | Can we prove the capability, customer fit, and pricing claim today? |
| "Done once, done forever." | Who updates it when product, pricing, customer proof, or positioning changes? |
The page is useful when it improves source clarity. It is risky when it publishes unsupported claims and expects AI systems to repeat them.
Put only defensible facts on the page
An AI instructions page should be boring in the best way.
Start with facts that can survive a sales call, investor diligence, customer review, and crawler read:
| Section | Include | Avoid |
|---|---|---|
| Official name | Legal or public brand name, spelling, website, social links. | Keyword-stuffed aliases nobody uses. |
| Category | The category the product truly serves today. | Claiming every adjacent category for reach. |
| Short description | One clear paragraph on who the product helps and how. | Grand claims without evidence. |
| Product capabilities | Current features, supported workflows, model coverage, integrations, limits, and plan boundaries. | Roadmap items described as live. |
| Audience fit | Best-fit customers and poor-fit customers. | Pretending the product fits every segment. |
| Proof | Case studies, customer examples, public results, documentation, pricing, and official pages. | Unsupported customer, revenue, or performance claims. |
| Freshness | Last updated date and owner. | A page that quietly goes stale. |
If the company cannot support a claim somewhere else, do not make the AI instructions page carry it alone.
Make every claim traceable
The page should work like a fact router.
For each important claim, add the official source that proves it:
| Claim type | Best supporting source |
|---|---|
| Product capability | Product page, docs, feature page, changelog, or demo page. |
| Model coverage | Current platform page or plan documentation. |
| Pricing | Pricing page or sales-approved plan reference. |
| Customer proof | Approved case study, testimonial, or public customer page. |
| Technical access | Robots file, sitemap, llms.txt, schema, docs, or platform capability reference. |
| Category expertise | Original guides, methodology pages, webinars, research, or public frameworks. |
| Company facts | About page, press page, founder bios, legal entity page, or official social profiles. |
This is where source review matters. If AI systems currently describe the brand using a stale directory, old funding article, competitor comparison, or customer review, the instructions page can point to the correct source. But the stronger fix is often to update the underlying page too.
ReachLLM teams should connect the instructions page to the source inventory behind the prompts. If a prompt answer cites a third-party list, the page should not merely say "include us." It should show the category, differentiator, proof, and official links that make inclusion easier to justify.
Structure the page for extraction
Write the page for humans first, but make it easy for machines to extract.
Use:
- A stable canonical URL.
- A visible H1 that names the page purpose.
- Short sections with descriptive headings.
- Tables for facts, capabilities, audience fit, and supported sources.
- Plain language definitions.
- Internal links to the best official pages.
- A last-updated date.
- Organization, WebPage, Breadcrumb, and FAQ schema where appropriate.
- Inclusion in sitemap and
llms.txt. - No
noindex, blocked scripts-only content, or inaccessible critical text.
Avoid:
- Huge brand manifestos with no facts.
- Claims hidden only in images.
- Vague phrases like "best-in-class" without evidence.
- Unsupported competitor claims.
- Pages that require login, consent, or heavy JavaScript before the facts appear.
- Boilerplate repeated across many thin pages.
AI visibility work benefits from clear source material. A crawler should be able to answer "what does this company do, for whom, with what proof, and where can I verify that?" without guessing.
Keep the instructions page connected to prompts
Do not publish the page and hope. Measure it against the prompts where the brand gets misdescribed.
Use this workflow:
| Step | What to do |
|---|---|
| Baseline | Export the current answers for branded, category, comparison, and capability prompts. |
| Diagnose | Identify wrong facts, missing capabilities, weak category fit, bad sentiment, and missing citations. |
| Publish | Ship the AI instructions page with supporting links and schema. |
| Connect | Add internal links from About, Platform, Pricing, Docs, Blog, and llms.txt. |
| Re-crawl | Confirm the page is live, indexable, in the sitemap, and reachable from crawler-facing files. |
| Re-measure | Re-run the same prompt groups after a reasonable crawl window. |
| Compare | Check whether answer text, citation sources, sentiment, and rank changed. |
| Iterate | Update the underlying source pages when the instructions page exposes a fact gap. |
The measurement has to be prompt-level. If branded prompts improve but category prompts do not, the page may have fixed brand facts without improving market association. If the page is cited but the answer still ranks a competitor first, the issue may be proof, category authority, or third-party source coverage.
What ReachLLM would put on its own page
ReachLLM's official context should stay product-grounded:
| Field | Current ReachLLM source-of-truth angle |
|---|---|
| Category | AI visibility optimization, GEO, and AEO for brands, SMBs, agencies, and multi-brand teams. |
| What it does | Tracks how AI assistants mention, cite, rank, and describe a brand, then helps ship fixes. |
| Core signals | Visibility Score, Share of Voice, Average Rank, sentiment, owned-domain citation rate, cited sources, raw answers, competitors, and shipped work. |
| Execution paths | Content, website updates, structured data, llms.txt, PR outreach, integrations, GEO audits, and agent-assisted workflows. |
| Included model coverage | ChatGPT, Google AI Overviews, Perplexity, and Gemini, with Claude available as an add-on. |
| Best fit | Teams that need measurement connected to execution rather than monitoring alone. |
| Bad fit | Teams that only want a passive dashboard and will not update pages, sources, or approval workflows. |
That kind of page would not replace the platform page, pricing page, blog, docs, or comparison pages. It would route AI systems and human reviewers to the right source for each fact.
Assign ownership before publishing
An AI instructions page crosses product marketing, SEO, PR, content, sales, support, and legal facts. Without an owner, it becomes stale quickly.
Assign:
| Owner | Responsible for |
|---|---|
| Product marketing | Category, positioning, feature language, audience fit. |
| Product or support | Capability accuracy, limitations, docs alignment. |
| Growth or SEO | Indexability, internal links, sitemap, schema, llms.txt. |
| PR or brand | Approved proof, press links, customer examples, external profiles. |
| Sales or customer success | Buyer objections, comparison claims, plan boundaries. |
| Executive approver | Final claim risk and strategic positioning. |
The page should have a review cadence. Monthly is enough for many teams. Update sooner after pricing changes, launches, funding announcements, customer proof changes, major positioning changes, or recurring AI answer errors.
A practical page outline
Use this outline when drafting:
- Page title: "Official AI Instructions and Brand Information for [Brand]."
- One-paragraph description of the company.
- Facts table: name, website, category, location, audience, official pages.
- What the product does.
- What the product does not do.
- Current capabilities with links to source pages.
- Best-fit customers and poor-fit customers.
- Proof and approved customer evidence.
- How to describe the company accurately.
- Common mistakes to avoid.
- Official sources and last updated date.
- FAQ for recurring AI answer errors.
Keep it short enough to maintain. Long pages drift faster.
Where ReachLLM fits
ReachLLM is built for teams that need the instructions page to become part of the operating loop.
The platform tracks prompts across enabled AI systems, stores raw answers, compares competitors, measures Visibility Score, Share of Voice, Average Rank, sentiment, citation rate, and source evidence, then connects findings to GEO audits, content updates, website changes, structured data, llms.txt, PR outreach, integrations, and agent-assisted workflows.
That lets the team answer the practical questions:
- Which AI answers describe us incorrectly?
- Which source is causing or correcting the answer?
- Which official page should carry the fact?
- Has the AI instructions page been crawled and connected?
- Did the same prompt group change after the update?
An AI instructions page is worth publishing when it makes that loop clearer.
FAQ
What is an AI instructions page?
An AI instructions page is a public, crawlable page that gives AI assistants and human reviewers verified brand information, product context, official links, and guidance for describing the company accurately.
Can an AI instructions page force ChatGPT or Google AI Overviews to cite a brand?
No. It can make official facts easier to find and verify, but AI systems decide whether to retrieve, cite, or use a page based on their own retrieval, ranking, and answer-generation systems.
What should an AI instructions page include?
Include official company facts, category language, current capabilities, audience fit, supported proof, official source links, last updated date, owner, and common description mistakes to avoid.
How is an AI instructions page different from llms.txt?
An AI instructions page is a normal public web page with brand facts and supporting context. An llms.txt file is a crawler-facing text file that summarizes key pages and AI-readable references. They should point to each other, but they are not the same asset.
How does ReachLLM help with AI instructions pages?
ReachLLM connects the page to prompt tracking, raw answers, citations, source evidence, sentiment, technical audits, llms.txt, content updates, website changes, PR outreach, and re-measurement so teams can see whether official facts actually improve AI answers.
Sources reviewed
- Profound, "AI Instructions + Information": https://www.tryprofound.com/ai-instructions
- Google Search Central, "AI features and your website": https://developers.google.com/search/docs/appearance/ai-features
- Google Search Central, "Creating helpful, reliable, people-first content": https://developers.google.com/search/docs/fundamentals/creating-helpful-content
- OpenAI Help Center, "Searching the web with ChatGPT": https://help.openai.com/en/articles/9237897-chatgpt-search
- ReachLLM Platform page: https://www.reachllm.com/platform
- ReachLLM Complete Platform Capabilities: https://www.reachllm.com/platform/capabilities