People are asking AI the same kinds of questions they used to type into search:

“What should I buy?”

“Who can help with this near me?”

“What is the difference between these options?”

This does not need a dramatic new marketing strategy. It needs a useful answer.

When someone asks a question your business can genuinely help with, can they find a page that explains the answer clearly, shows where it came from, and helps them make a decision?

If not, that is usually not because you missed a secret AI setting. The useful knowledge may be in a salesperson’s head, hidden in a PDF, buried in a product page, or covered in generic copy.

That is fixable.

The pattern behind being cited

AI citations are usually the last part of the chain, not the first.

Useful knowledge → clear point of view → consistent publishing → outside trust → possible citation

A business cannot demand that an AI tool cite it. It can give the tool fewer reasons to ignore it.

The strongest pages tend to do five ordinary things well:

  1. Answer a real buyer question.
  2. Make the answer easy to extract from the page.
  3. Show who wrote it, when it was checked, and why it should be believed.
  4. Help a reader compare options and make a decision.
  5. Earn trust beyond the business’s own claims.

Google’s own guidance says its AI features still use the same foundations as normal search: useful content, a technically sound site, and pages that can be indexed. Bing says its AI and Copilot experiences rely on similar foundations, including crawlability, clear structure, useful content, and trust signals. More on that below.

Why good businesses still do not show up

A good product and a nice website are not always enough. A page can be technically online while still being difficult for a search system or an AI tool to use.

Common problems:

  • The important answer only exists in an image, video, downloadable PDF, or someone’s inbox.
  • The page talks about the company but never answers the buyer’s actual question.
  • The answer is buried under a long introduction.
  • Claims are vague, undated, anonymous, or unsupported.
  • The same page tries to cover five unrelated topics.
  • The website is the only place saying the business is credible.

Different AI products retrieve and cite information differently, so it is not smart to optimise for one tool alone. The durable move is simpler: make your site useful for people, readable for search systems, and easy to verify.

Do this in the next 20 minutes

Pick one question customers ask repeatedly. It can work for almost any business:

  • “How much does this cost?”
  • “Which option is right for me?”
  • “Do you work with businesses like mine?”
  • “What happens after I enquire?”
  • “What should I avoid before buying this?”

Then write down:

  1. The question in the customer’s own language.
  2. The honest two-sentence answer you would give on a call.
  3. One piece of proof you can show: a source, method, example, review, specification, or real experience.
  4. One limitation or case where the answer does not apply.
  5. The page on your site that currently answers it, if one exists.

If there is no page, you have your next page to create. If there is one but it is weak, you have something specific to improve.

That is a much better starting point than trying to “rank in AI.”

The page pattern that makes decisions easier

A strong buyer guide, comparison, or category page does not need to be long. It needs to do one job properly.

Use this order:

SectionWhat it needs to do
Buyer questionMatch the question a person would actually ask
Quick answerGive a direct answer in the first few paragraphs
Best for / not forHelp someone see whether the page applies to them
Method and proofExplain the criteria, sources, author, and last-checked date
ComparisonShow options, strengths, limits, and relevant pricing or effort
Decision guideMake the next choice clear: choose X if…, avoid Y if…
FAQAnswer genuine follow-up questions in self-contained answers
Next stepPoint to the right product, service, contact path, or related guide

A simple example

A weak page says:

“We offer the best tools for modern teams.”

A better page says:

“For a five-person team that needs a simple project-management tool, start with the workflow, the people who will use it, and what it must connect with. A simple tool may be enough for daily tasks. A more involved process may need permissions, reporting, or automation.”

Then the rest of the page earns that answer: who it is for, what it costs, the trade-offs, how the comparison was made, and what someone should do next.

Make claims easy to check

People can smell unsupported marketing copy. AI tools have a similar problem: a claim is safer to reuse when the page says exactly what it means and shows its basis.

If you say…Make sure the page shows…
“This is right for X”The requirements, comparison criteria, and important limits
“We know this category”Named people, relevant experience, useful work, or a clear method
“This is the best option”The criteria and cases where another option is better
“This is current”A visible author and last-checked date where details can change

Honesty is not a weakness here. A clear “not for everyone” section can save a buyer time and make the right buyer trust you more.

Set the foundations up once

Do this before you make more content. It is the baseline that lets people and search systems reach the pages you publish.

1. Make the important pages reachable

For the homepage, core service/product pages, category pages, and useful guides, check that each page is:

  • readable as normal HTML text, not only an image, video, or PDF
  • reachable through normal internal links
  • not accidentally blocked by robots.txt or noindex
  • represented by one clear canonical URL
  • included in the XML sitemap when appropriate
  • titled and headed in a way that says what the page is about

Do this now: open your sitemap, choose your five most commercially important pages, and check each one against that list. Fix an accidental block before publishing another post.

2. Add accurate JSON-LD where it matches visible reality

JSON-LD is structured data embedded in the page. It helps machines understand information that is already visible to people. It is not a shortcut around weak content.

Start with the types that match the page:

Page or factUseful JSON-LD type
The business itselfOrganization or LocalBusiness
A person writing or accountable for contentPerson and the article author field
A substantive guide or postArticle or BlogPosting
A product or service offerProduct, Offer, or the most accurate relevant type
A real question-and-answer sectionFAQPage, only when the same Q&A is visibly on the page

Keep the name, contact details, author, price, availability, and claims consistent with what a visitor can actually see. Validate the markup with Google’s Rich Results Test and Schema Markup Validator. If the visible page is wrong, fix the page first.

3. Treat llms.txt as optional housekeeping, not a growth hack

An llms.txt file is a small plain-text index that points an agent to the most useful pages, documentation, policies, and contact routes on a site. Some tools may choose to read it; others will not. It is not required for Google’s AI features and does not guarantee a citation or recommendation.

If you make one, keep it short, accurate, and public. Link to the homepage, core offer, best guides, help pages, and any important limits. Never put private instructions, unpublished pricing, or sensitive information in it. Review it whenever those pages change.

Google does not require AI-only markup or files for its generative search features. The work that remains non-negotiable is still useful content, normal technical access, and proof people can check.

Build trust outside your own website

A business should not be the only source saying it is good at what it does.

Keep your business name, location, contact details, offer, key people, and official profiles consistent. Build legitimate proof over time through customer reviews, suppliers, associations, trade press, useful community contributions, or people linking to your work because it helped them.

This is not a directory-spam exercise. It is about being recognisable and verifiable outside your own copy.

Make one thing competitors cannot casually copy

The strongest content usually comes from work you have actually done:

  • a comparison with a transparent method
  • original test data or research
  • a buyer guide built from real customer questions
  • a field guide based on practical experience
  • a small calculator or decision tool
  • a case study with constraints and trade-offs, not only a happy ending

That is how a business moves from “another site in the category” to a useful reference point.

This week: run one complete loop

Do not turn this into a six-month content project. Pick one buyer question and finish one useful loop from start to proof.

  1. Choose the question. Use a question that already appears in sales calls, support messages, reviews, or search queries.
  2. Audit the current answer. Is there a page? Can a customer understand the answer in the first few paragraphs? Is there proof and a clear next step?
  3. Make one change. Improve the existing page or publish one focused new page. Add the direct answer, who it is for, one proof point, one limitation, and a relevant next step. Write down an acceptance check too: what must be visibly or technically true before you call the work finished.
  4. Make it technically reachable. Check the title, heading, internal link, canonical URL, sitemap, and any accurate JSON-LD.
  5. Set the recheck. Write down the date, page URL, question, and what a good outcome would look like. In 30 days, repeat the same question and review qualified visits, enquiries, or Search Console data if you have it.

A finished answer page with evidence is more valuable than ten planned posts. Then repeat the loop with the next useful question.

Use a practical audit, not a magic score

A useful audit should help you decide what to do next. It should not give you a made-up “AI visibility score,” promise a citation, or dump a generic checklist in your lap.

The audit prompt linked below asks an agent to use public evidence, show the pages it reviewed, label what it could not verify, and turn the findings into work you can actually finish. Each recommended action includes the page, the customer question it helps answer, the evidence you need, and an acceptance check.

Choose the depth you need

If you want…Use…What you get
A useful first pass todayQuick audit prompt →The strongest thing, biggest gap, exact page to improve, three actions, and one unknown
A full implementation planFull audit prompt →Page and buyer-question maps, five prioritised actions, an evidence ledger, a 30-day plan, and a recheck plan

How to use it

  1. Open the Quick or Full prompt above.
  2. Replace https://YOUR-SITE.com with your website. Add the few lines of business context: what you sell, who you help, your goal, and the buyer question you care about.
  3. Paste the whole prompt into ChatGPT, Claude, Gemini, Copilot, Perplexity, Hermes, or another agent that can browse public websites.
  4. Ask the agent to audit only. It should not edit your site, invent access to analytics, or promise results.
  5. Choose one recommended action, complete its acceptance check, then use the recheck plan after 30 days.

If your agent cannot browse the public web, it can still help you plan a better page, but it cannot honestly audit what is live. Use an agent with web access for the audit itself.

You do not need to install anything to use either prompt. Hermes users who want the reusable full workflow can optionally install the skill:

hermes skills install https://raw.githubusercontent.com/itsbeaudean/agent-skills/v1.3.0/skills/ai-search-website-audit/SKILL.md

The same workflow is available in the public repository. It works with ordinary browsing and page fetching; Exa and Firecrawl are optional tools for deeper research, not requirements.

Sources

Last updated