AI visibility, representation and integrity are not the same thing
A monitoring tool records that the Brand was mentioned. The team celebrates an AI-visibility win.
Nobody checks whether the answer described the right Product, preserved the qualification or directed the customer to a current next step.
| Measure | Question |
|---|---|
| AI Visibility | Did the organisation appear? |
| AI Representation | What did the answer say, omit or imply? |
| AI Narrative Integrity | Was the representation accurate, current, supportable and appropriate to the decision? |
Visibility may coexist with outdated Product facts, missing conditions, source confusion, unsupported comparison claims, wrong locations or people, obsolete next steps, incomplete context, and overreliance on third-party sources.
Record the customer question and prompt, platform and date, answer and links or cited sources, material claims, accuracy and omissions, Product and entity identity, qualifications and context, likely decision consequence, evidence status, and owner and next observation.
A single answer is an observation. Repeated, comparable observations may support a stronger pattern. The organisation should not turn one favourable or unfavourable output into a universal claim.
Mention or inclusion rate, Product/entity inclusion, linked or cited source frequency, share of relevant answer environments.
Material facts included and omitted, Product and entity accuracy, comparison framing, next-step accuracy, source type used.
Accuracy, consistency, provenance, currency, context preservation, entity coherence, technical accessibility.
The three groups should not be collapsed into one vanity score.
A favourable mention may still fail the customer when the Product is not suitable for the stated need, the fee excludes an important condition, the answer links to the wrong region, the next step has changed, the source is outdated, or a regulated qualification is omitted.
Use a controlled observation set: defined customer questions, consistent platform conditions where possible, dated captures, comparable locations and repeated observation over time. Variation should be part of the result, not removed from the record.
The finding becomes material when the representation can plausibly affect an important customer decision, Product truth, reputation or commercial path. A low-impact wording difference may require no action. An inaccurate eligibility condition close to application may require urgent ownership.
These controls prevent AI reporting from becoming another visibility dashboard that leaves the important interpretation to the reader.
Measure what the answer means—not only whether it exists.
Explore AI Narrative Integrity, or return to the cornerstone framework.
Govern technical conditions affecting search eligibility, first-page presence, customer…
See how public sources, entity consistency, evidence provenance and AI representation shape…
See how SI governs baselines, contribution, value and limitations, with populated…