When competitors appear in AI recommendations and your business does not, start with evidence rather than a promise to rank. Collect the questions customers asked, inspect which sources were cited, verify crawl access and compare how clearly each business explains services, locations, proof and limitations.
This is an SEO and technical SEO investigation. AI answers can vary by wording, location, time, available web results and product mode, so one screenshot is a lead for diagnosis, not a stable ranking report.
Capture the customer evidence precisely
Ask customers for the question they entered, approximate date, product, whether web search was used, location context and the response or sources they saw. Remove personal and confidential information before storing examples.
Group prompts by decision: finding a provider, comparing options, validating trust, checking local availability, estimating cost or learning a process. Record which competitors appeared and whether the answer linked to their website, an independent publication, a directory, review source or no visible citation.
Use an AI recommendation evidence log to preserve the prompt and source context.
Do not ask employees to repeatedly click or manipulate results. The purpose is to understand coverage and evidence, not manufacture engagement.
Verify OpenAI search crawler access separately
OpenAI's official crawler documentation states that OAI-SearchBot is used to surface websites in ChatGPT search results. Sites that opt out will not appear in ChatGPT search answers, though they may still appear as navigational links.
Check robots.txt, CDN, firewall, bot protection and server responses for OAI-SearchBot. Allow requests from OpenAI's published searchbot IP ranges when appearing in ChatGPT Search is desired. Verify with server logs and a controlled fetch rather than assuming a robots rule reaches the application.
OpenAI documents OAI-SearchBot and GPTBot as independent controls. A business can allow the search crawler while disallowing GPTBot for model-training use. Make that choice deliberately with technical, legal and policy owners.
ChatGPT-User is used for certain user-triggered actions and is not the control for automatic Search crawling. Do not confuse an occasional user-agent request with sustained Search eligibility.
Check important pages return a successful status without authentication, geoblocking challenges or consent walls that hide essential content. Ensure text and links exist in rendered, accessible HTML.
Audit conventional search accessibility too
AI search visibility still depends on content being available and understandable. Crawl the site, inspect indexability, canonicals, redirects, internal links, sitemaps and server rendering.
Identify service pages or location pages that exist only in client-side interactions, duplicate pages that declare conflicting canonicals, and content hidden inside images or downloadable files without an accessible web equivalent.
Check whether the site blocks important bots globally while allowing a user-agent string in robots.txt. A firewall can still deny the request. Monitor response codes and content size by legitimate crawler.
Repair broken or redirected internal links and ensure important pages are reachable through logical navigation. Search systems and users both need stable paths.
Make the business entity unambiguous
Publish a clear organisation page with preferred name, alternate or legal name where relevant, service scope, locations, contact, leadership or accountable expertise, and official profiles. Explain relationships between a parent company, trading brand and product.
Use Organization or LocalBusiness structured data only for facts visible and accurate on the page. Google's Organization structured-data guidance explains that properties such as name, alternateName, URL, logo, address, telephone and sameAs can help disambiguate an organisation.
Keep names, domains, contacts and location details aligned across the official site and claimed profiles. Legitimate formatting variations are acceptable, but contradictory names, closed addresses and old service descriptions weaken clarity for customers and machines.
Create dedicated location pages only when the location has genuine operational information. Do not create hundreds of near-identical city pages to simulate presence.
Compare answer coverage, not keyword count
For each customer prompt, compare your useful coverage with the sources that were surfaced. Does the site clearly answer the question? Does it explain who the service suits, the process, constraints, evidence, price drivers and next step?
Google's guidance on helpful, reliable, people-first content recommends content for an intended audience that demonstrates experience and leaves readers feeling they learned enough to achieve their goal. Those principles are also a strong defence against thin pages created only for AI queries.
Build one substantial page for a real decision instead of many slight prompt variations. Use a direct answer, structured headings, comparison dimensions, original examples, limitations and sources.
Show firsthand experience through methods, artefacts, measured outcomes and accountable authorship. Replace vague claims such as leading provider with evidence a customer can evaluate.
Update volatile information and show review dates where useful. An old service description can make a correct exclusion more likely.
Strengthen proof and corroboration
Recommendations may rely on sources beyond the company's own claims. Inventory accurate profiles, professional associations, partner directories, customer reviews, case studies, credible editorial coverage and public records relevant to the service.
Correct factual inconsistencies in sources the business controls or can claim. Follow each platform's correction process. Do not create fake reviews, paid editorial disguised as independent evidence or networks of low-quality citations.
Ask customers for honest reviews without dictating language or conditioning service. Obtain permission for case studies and describe context, intervention, result and limitations.
Provide journalists, partners and directories with a concise fact sheet linked to the authoritative website. Track effective dates and retire old logos or descriptions.
Corroboration should reflect real reputation and expertise. A large volume of manufactured mentions can create legal, platform and trust risk.
Publish content that can be cited responsibly
Make key facts easy to locate and connect them to evidence. Use descriptive page titles, clear authorship, dates, tables with labelled units and links to primary sources.
For comparisons, state the decision context and use consistent criteria. Do not create competitor attack pages or claim superiority without evidence. Honest situations where another approach fits better can increase usefulness.
For local and regulated advice, identify geography and limits. Cite current official sources and distinguish general information from professional advice.
Use crawlable text for the primary answer. Video, images and PDFs can add evidence, but provide captions, transcripts or web summaries where they carry essential information.
Avoid schema that does not match visible content. Passing a validator does not make an unsupported claim true or guarantee inclusion in an answer.
Design a reproducible prompt benchmark
Build a set of 20 to 50 prompts from real customer language across discovery, comparison, trust and local intent. Record prompt, expected relevant facts, unsafe claims to avoid, product mode, date and location context.
Run the set on a documented cadence, not repeatedly until a desired answer appears. Capture whether the business is mentioned, accurately described, cited and linked, along with competitor and source patterns.
Score dimensions separately:
- relevant mention;
- factual accuracy;
- citation or source presence;
- service and location fit;
- appropriate caveats;
- visibility of competitors;
- answer stability across runs.
Treat results as samples. Do not publish a single percentage as universal AI share of voice without explaining the prompt set and variation.
Connect visibility to business outcomes
Monitor referral sessions from identifiable AI and search sources, landing pages, qualified actions and customer-reported discovery. Attribution will be incomplete, so combine analytics with form and sales questions.
Add a concise how-did-you-hear-about-us question only when it supports a decision. Preserve free-text answers alongside controlled categories so new sources can emerge.
Measure accuracy and fit, not mention volume alone. An AI answer that recommends the business for an unsuitable service can create wasted sales work and customer disappointment.
Track branded demand, relevant citations and source coverage over time. Annotate major content, technical and profile changes. Avoid claiming causation when several signals changed together.
A 60-day response plan
During week one, collect customer prompts, capture sources and verify OAI-SearchBot plus conventional crawl access. Fix accidental blocks and server errors.
During weeks two and three, reconcile organisation, service and location facts across the website and high-value profiles. Correct duplicates and obsolete information.
During weeks three to six, improve the highest-priority decision pages with original evidence, direct answers and current sources. Consolidate overlapping content.
During weeks six to eight, run the benchmark again, monitor qualified referrals and identify which source gaps remain. Continue improving customer value rather than producing bulk variants.
My take: become the clearest defensible source
My take is not to optimise for a mysterious recommendation algorithm. Make the business accessible, unambiguous and well evidenced for the questions customers actually ask. Earn corroboration and measure the result honestly. There is no guaranteed position, but these improvements strengthen search, AI visibility and customer trust together.
Frequently asked questions
Can a business pay to rank in ChatGPT recommendations?
Do not assume an organic recommendation can be bought. Treat paid placements, where offered and labelled, separately from earned search visibility.
Should we allow GPTBot to appear in ChatGPT Search?
OpenAI documents GPTBot and OAI-SearchBot as independent controls. OAI-SearchBot is the relevant automatic search crawler.
Does adding more schema guarantee a mention?
No. Accurate structured data can help clarify entity facts, but it does not guarantee ranking, citation or recommendation.
How quickly should visibility change after a fix?
Technical access changes can be processed faster than reputation and source coverage. Measure over time and avoid promising a fixed result date.
Build search visibility on evidence
If your business needs technical and content work for AI-era discovery, get in touch for SEO and technical SEO.
Related guides
For further background, read our ai search landscape and ai search optimization guide. The ai visibility glossary explains the terminology used in those guides.