A business missing from one AI answer is a starting observation, not a diagnosis. Keep the exact question, engine, country setting, date and source links. Then distinguish whether the engine found the wrong business, could not access useful evidence, or simply recommended someone else.
Four different problems need different evidence
Scroll sideways to read every column.
| Observation | What to inspect | Possible work |
|---|---|---|
| Wrong company or founder | Branded answer, exact domain and cited identity pages | Consistent visible company/founder facts and authentic external corroboration |
| Useful public pages cannot be retrieved | Access/index evidence for the particular engine; readable content | Technical correction when a specific barrier is established |
| Correct entity, weak substantiation | Service claims versus cited pages and client-side acknowledgement | Specific service/proof pages and verified engagement evidence |
| Correct entity, not shortlisted | Non-branded buyer question, alternative recommendations and associated sources | Evaluate buyer fit, evidence gaps and repeated observations; inclusion cannot be promised |
Use a branded control without calling it discovery
Ask a separate control question naming the company and exact domain. That can expose namesake confusion or inaccurate role descriptions. Keep it outside the denominator for non-branded discovery questions: an answer finding a business you explicitly named is different from a buyer discovering it.
Strata’s retained branded controls found Dean and the offer, but some blurred diagnosis with recovery or described approved client names inaccurately. That supports improving factual clarity and verification. It does not show that Strata was recommended for non-branded hiring questions.
Repeat the actual buyer question
Keep the wording fixed and repeat on separated dates where feasible. Record account/login or provider-model conditions as unknown when they are not disclosed. Save failures separately; a timeout is not “brand absent.” Compare each engine independently and retain inconvenient results.
Check the sources supporting a recommendation. A software vendor can be misdescribed as an agency, or a case statistic can be repeated with the wrong causality. Inspect whether the cited page actually establishes the service, client role, period and result claimed.
Improve useful evidence before chasing answer formats
Make the offer, implementation responsibility and relevant case evidence clear on accessible pages. Align founder/company facts and seek authentic client acknowledgement where appropriate. Those are sensible technical and entity practices; they do not guarantee a citation or recommendation.
For Google, ordinary Search eligibility applies to its AI features; the guidance does not require a special AI schema. OpenAI distinguishes its search crawler from its training crawler, so a training preference is not interchangeable with a search-access decision.
Engine-specific primary sources: Google’s AI-feature guide and OpenAI crawler roles. Neither source proves a universal cross-engine growth prescription.
Keep visibility and commercial contribution separate
JS Wallmatrix’s dated AI Overview evidence shows citations for selected buyer questions. It does not prove that AI caused its enquiry cohort. An observed referrer and a buyer’s reported influence also answer different questions and can overlap.
Use the AI-enquiry measurement guide and ungated recording template to connect accepted requests with qualification and opportunities. An engine mention alone is not a lead. For practical scope, see AI-search optimization and indexing/access diagnosis.