THE METHOD & ITS LIMITS
Every edit starts
with evidence.
We investigate how your product appears in specific buying answers and where the public evidence could be clearer. We cannot see inside a model’s decision process.
01. Observe an actual buying answer.
Agree twelve buyer questions and take two search-enabled samples per question. Unbranded discovery questions do not name your company or restrict results to your domain. We preserve exact prompts, model, timestamps, answers, citations and errors.
The pilot uses the OpenAI Responses API with web search. An answer without a completed search call is not counted as a valid search-enabled observation. These samples do not represent personalized ChatGPT sessions, other AI engines or universal rankings.
02. Check whether the gap is meaningful.
We inspect up to five agreed public pages, three competitor homepages and relevant returned sources. An omission alone is not a defect. We compare the buying intent, product fit, competing options and the accuracy of the answer.
Repeated omissions, legacy framing or inaccurate comparisons can justify further investigation. Small wording differences do not establish commercial harm. We do not turn a text match into a recommendation rate or a visibility score.
03. Trace claims to public sources.
The source / claim map separates what a page says, what the AI answer says, and our interpretation. Citations show sources returned with an answer; they do not reveal every influence on that answer or prove why a product was excluded.
A missing heading, old description or inconsistent category is a potential repair opportunity. It is not proof of causation. Unreadable or restricted pages are reported as such, never filled with invented findings.
04. Prepare three reviewed page changes.
Each package identifies the page and placement, the issue, supporting evidence, exact proposed copy and implementation checks. A human reviewer checks the claims and source relevance before handoff.
Your product team validates business facts and approves the final wording. Your web/content team implements and publishes. No invented capabilities, customer proof or performance claims; no requirement to give us CMS access.
05. Repeat, with the limits visible.
After publication, repeat the documented questions within the agreed 30-day pilot window. Keep the provider, model and search settings comparable and disclose any unavoidable changes.
Two samples can be correlated. A before/after difference does not prove that page edits caused it. Sources, retrieval and models change independently. No inclusion, ranking, traffic or revenue outcome is guaranteed.