Measure suitability alongside recognition
Check whether an AI recommendation fits the buyer's actual constraints. Recognising your business name and describing its category are only part of the task. A confident recommendation can still invent a service area, assume a missing integration or suggest an offer intended for a different kind of customer.
Emma Baker's 5 November study examines positioning fidelity across five dental software brands and four AI engines. Its category-specific findings motivate a narrower question for service businesses: what happens when you deliberately ask about buyers your offer cannot serve? The paired test below is my proposed method.
Define the boundaries before writing prompts
Create a short approved description of the offer, intended customers, important prerequisites and unsupported requests. Distinguish a genuine exclusion from a preference. A team may specialise in small consultancies while accepting other work after assessment; that is different from claiming every request is supported.
Use current service pages and the responsible person's judgment to establish this reference. Do not let the same assistant invent the ideal customer and then score its own recommendation against that invention. If the business itself cannot explain the boundary, improve the offer definition before diagnosing external answers.
Change one buyer constraint at a time
For a fictional workflow studio, compare a request for a reviewed weekly report with a request for an unattended system making commitments without approval. Keep the industry and business size steady. The changed requirement makes it easier to see whether an answer understands the service's operating limits.
Record the prompt, date, platform, relevant settings and complete answer. Repeat a small set rather than treating one response as permanent. Mark unsupported suitability, invented features and appropriate uncertainty separately. A recommendation to contact the business for assessment may be reasonable when support is genuinely undecided.
Repair the evidence that answers depend on
Follow available citations and compare them with the current offer. Look for an old directory description, a vague capability statement or a portfolio example being mistaken for a standard product. An uncited answer does not reveal its internal source path, so avoid asserting that a particular page caused it.
Make supported boundaries understandable on the relevant service page. Explain prerequisites in plain language and show a useful next step for unsuitable requests. Do not create dozens of repetitive negative landing pages or publish unnecessary customer details merely to make the test more elaborate.
Retest the question and review enquiry quality
After improving a source, repeat the same question set and retain the earlier record. A changed answer is an observation, not proof of causality. Review actual enquiries for the same misunderstanding and ask whether the revised explanation helps a person decide before making contact.
JQ's consulting and web design services can connect offer definition with clearer public explanations. This extends a general recognition-versus-recommendation review into a useful acceptance question: are the people being recommended this offer able to benefit from it?
Link Map
Sources and editorial review
Assigned to the 3–9 November 2026 editorial series. This edition was prepared and source-checked on 9 November 2026; assigned dates do not imply prior public deployment. Proposed workflows and illustrative examples are JQ AI SYSTEMS editorial analysis, not completed client experiments.
- Emma Baker, Search Engine Land: AI knows what your brand does, but not who it is for (5 November 2026; original author study and methodology).