How to help AI assistants find and describe your business
Make your business easy to identify, your offer easy to understand and your claims easy to verify. Then measure mentions, citations and actual enquiries separately. AI visibility builds on useful public information and technical access; no file or formatting trick guarantees a recommendation.
- Fit
- Who the service is for
- Evidence
- What supports the claim
- Next step
- How to check scope or contact you
Understand what being found can mean
AI visibility can mean several things: an assistant names the company, links to a page, describes the offer correctly or recommends it for a particular need. Record these separately. A business may be mentioned in a comparison for an unsuitable service, or its page may be cited for a general explanation without any buying intent. Counting every appearance as a sales opportunity obscures the useful information.
A mention is not a recommendation
Different products and modes can use different sources, retrieval methods and presentation. Do not assume that a successful test in one assistant describes every AI search experience. Even the same question can produce different answers on different occasions. A visibility programme should specify the products, languages, questions and dates being tested, and describe the results as observations rather than permanent rankings.
A citation is not a customer
Start with the business outcome: qualified enquiries from people who understand the offer. An accurate answer may help that outcome even without an immediate click, while a visible citation may bring visitors who do not fit. The website still needs a persuasive service page and working contact path. Improving source material is useful work; proving its commercial effect requires more than a screenshot.
Choose buyer questions that deserve a useful answer
Collect real language
Use actual enquiries, sales conversations and available search data to collect the way customers describe the problem. Separate a request for a supplier from a request for instructions or a price comparison. For example, Webflow agency for an architecture studio has a different purpose from how to change a Webflow heading. Both can be valid content topics, but they need different answers and expectations.
Group by decision
Group related questions around the decision they support. A service page can explain suitability, deliverables, price factors and the first step without creating a separate URL for every wording. A guide can compare approaches or help prepare a brief. A case study can show specific work. Connect these pages so a reader can move from research to checking whether the agency fits.
Prioritise relevance and evidence
Prioritise topics where you offer the service and have something useful to contribute. Search suggestions and the presence of results are clues about language, not measured search volume. When demand is uncertain, label it as a hypothesis and monitor it. Avoid producing a large set of nearly identical city or technology pages merely because the keyword combinations are easy to list.
Fix access and eligibility before adding special files
- Public pages that can be read
- Check that important information is available in readable page content and that the intended public URL loads correctly. Review accidental login barriers, broken internal links and missing content in the rendered page. Google states that its AI search features use the usual Search foundations; there are no additional technical requirements beyond eligibility for Search and a snippet. Eligibility still does not guarantee inclusion.
- Crawling and indexing controls
- Review robots.txt, noindex directives and canonical URLs according to their actual purposes. Google explains that robots.txt controls crawling and is not a reliable way to keep a URL out of search results. Do not use it to protect private information. Check each provider's documented controls before changing crawler access, because a broad allow or block decision may affect more than the intended use.
- Structured data that matches the page
- Where relevant, structured data can describe information already present on the page, such as an organisation or article. It should agree with the visible facts and the supported format. Markup does not create evidence for an invented rating, office or price. Google does not require a special AI schema or an AI text file for its AI features; an optional facts file is not a substitute for useful pages.
Write pages that answer the decision in enough detail
A visitor looking for a supplier needs to understand what will be done, for whom, under what conditions and how the work starts. Open with that answer, then explain the choices that affect scope. Include a price or price factors when you can state them accurately. Explain limitations in normal language. A sequence of impressive slogans leaves the reader to assemble the actual offer.
Make the offer concrete
For a website service, describe the editing workflow, content responsibilities, integrations and handover. For automation, name the inputs, permitted actions and exception path. Use meaningful examples rather than repeating the service name. A comparison should help the buyer recognise when an option fits and when it does not. These are editorial recommendations for making the page useful, not a claim about a secret assistant ranking formula.
Make the evidence inspectable
Show real work with the client's permission and explain what the project demonstrates. Separate delivered functions from measured business outcomes; a CRM connection is not proof of increased revenue. Give research a method, date and limits. Google advises people-first content and does not prescribe a preferred word count. Use the length needed to answer the reader's questions, with structure that makes those answers easy to find.
Build consistent facts and credible references outside the site
Keep the business identity consistent
Check the company name, public contact, service descriptions and actual location across profiles you control. Correct contradictory or obsolete information. A remote agency can serve a market without claiming an office there. The practical value is that a prospective client can verify the same business through several places, instead of finding different offers and contact details that make the relationship uncertain.
Publish references with substance
A client project credit or a published account of the collaboration can be useful when the client wants it and the content explains real work. A short, accurate description with a relevant link helps a reader understand who delivered the project. There is no need for an exaggerated praise article. Do not invent testimonials, outcomes or editorial independence to manufacture supporting evidence.
Avoid links created only to manipulate rankings
Keep credits visible, proportionate and agreed with the site owner. Google lists hidden links and manipulative link schemes among its spam concerns; its policy also addresses paid links that pass ranking credit. Do not treat a particular number of links or a slow posting schedule as a guarantee of safety or impact. Prefer references that have a genuine purpose for people using the publishing site.
Measure mentions, referrals and enquiries separately
- A repeatable observation set
- Choose a manageable set of buyer questions and record the exact wording, language, product, date and relevant settings for each check. Save whether your business appears, which URL is cited and whether the description is correct. Repeat enough to notice variation. Do not present one favourable answer as a stable result or a sampled mention rate as a share of the whole market.
- Search and website evidence
- Use website analytics to inspect identifiable referral visits and landing pages, while recognising that not every influence produces a trackable referral. Google's documentation says traffic from its AI features is included in the overall Web performance reporting in Search Console. That report therefore should not be relabelled as a separate count of AI mentions. Compare each data source according to what it actually measures.
- Qualified business outcomes
- Connect enquiries to the page and source when those details are available, and ask how a prospect found you when attribution is unclear. Review whether the enquiry matches your services, budget and delivery model. A smaller number of relevant conversations can be more useful than a large increase in generic traffic. Keep test submissions and your own research visits identifiable so they do not distort the assessment.
Plan the first improvements in a useful order
A focused first work package
Select the services you want to sell, the buyer questions they answer and the pages that should support them. Check access and factual accuracy first. Then improve the weakest important page with a concrete offer, supporting work and a clear next action. Record a baseline of search performance, relevant AI observations and enquiries before publishing so later comparison has context.
Next, correct inconsistent external profiles and identify legitimate references to completed work. Review the contact path by submitting a clearly marked test and checking who receives it. A page that gets a useful citation but loses the enquiry at the form cannot meet the business goal. This work connects discovery with the actual sales process rather than treating AI visibility as an isolated score.
Review the evidence regularly and choose the next improvement from actual gaps: an incorrect fact, a buyer question without a useful answer, an inaccessible page or an unclear offer. No responsible scope can guarantee that an independent assistant will recommend a business. A good scope can specify the pages, technical checks, evidence and measurement that will be delivered, along with the limits of what can be observed.
How we can help
Our part
HeadPills can audit public facts, service content, technical access and a defined set of AI search observations. The work is scoped around relevant buyer questions and a functioning path to enquiry, with no promise of guaranteed recommendations.
Questions
Questions from readers
Do I need llms.txt to appear in Google AI features?
No. Google says its AI features do not require special AI text files or new markup. An optional facts file cannot replace accessible, useful public content.
Can you guarantee mentions in AI assistants?
No. We can define and deliver content, technical checks and a repeatable measurement process, but independent systems control their own answers.
Should we create a page for every wording of a query?
No. Group related questions by the decision they support. Separate pages should offer distinct value and correspond to real services or useful research tasks.
Make the offer easier to find and verify.
Share your site and priority services. We can assess the content and technical gaps that affect discovery.