What happened overnight?
This is an illustrative scenario, not a customer account. The people and the company are invented; every event in it maps to a real mechanism on ftes.ai, named as it happens.
The company is a business software vendor with a marketing lead and no dedicated content team. The Search Visibility FTE ran its weekly sample: each question in the approved prompt set asked of each selected engine, the answers recorded with the date and whether the site was cited. From the sample and the connected console it produced two candidate topics, each with a worth-it verdict: is the question asked, does the site answer it, who is cited, can it be answered from approved sources. It drafted the higher-verdict topic from last week's approval through the fact gate, which blocked one claim, a statement about an integration, because no approved source supported it.
Its detections fired once: for a question the site was cited on in every previous sample, a competitor is now cited on two engines and the site on one.
What does the marketing lead see?
The morning brief, first person: the sample taken, the two verdicts waiting, the draft waiting with one blocked claim, the detection, the pull request open from last week, and spend against budget. The lead opens the verdicts. The first is declined with a reason: the question is asked, but the site has a page that nearly answers it, and the lead prefers to rewrite that page, which the verdict's second question had surfaced as an option; the reason goes to the log and the next verdicts weight rewrites more. The second is approved and goes to the pipeline for drafting.
The detection is read as a signal: the competitor's page is opened, the lead notes what it answers that the site's page does not, and the gap becomes a candidate for next week's verdicts.
How is the blocked claim resolved?
The draft in the Approvals tab shows the fact gate's result: every claim with its source, and one claim highlighted as unsourced. The lead knows the integration exists and knows where it is documented, so the fix is to add the documentation page to the approved sources in the Setup tab, not to add a citation to the draft by hand. The FTE re-checks the draft against the updated sources, the claim passes, and the lead reads the draft for voice, sends it back once for a paragraph that sounds more like a competitor than like the company, and approves the revised version an hour later.
Nobody added a source after the fact. The source list grew, which is the mechanism working.
- Blocked: one claim, no approved source
- Fixed: source added to the approved list, draft re-checked
- Sent back once: tone, then approved
How does the page publish?
The site is built as code, so last week's approved page publishes as a pull request in the repository, opened by the FTE through its scoped connection with the content, the metadata, and the structured data in the format the engineers agreed at go-live. An engineer reviews it in the afternoon like any other change, asks for one heading to be shortened, which the lead does through the FTE with a send-back, and merges it. The publish is logged with the approver's name and the sources every claim traced to. Publishing stays at Level 0; the lead has no plan to promote it.
The library's piece on publishing to a Next.js repository describes the pull request route; for other stacks the endpoint package does the same job.
What does the lead do with the afternoon?
Reads the Performance tab: citation presence per engine as a range over the last four samples, console impressions for the target queries, pages published that passed the gate, and the unmeasured list, which includes attribution to sign-ups, stated plainly. Opens the Library tab and marks two older pages for rewrite to the answer-first discipline, which will appear as verdicts. Updates the Brand Brain with a corrected product fact the send-back revealed, with today's date as provenance. Then does the work only a person does: the positioning conversation with the founder about a new market.
The FTE did not attend that meeting. Its next sample runs in a week.
What does the record show at the end of the day?
The audit trail: the sample with its date and engines, both verdicts with the lead's decisions and the declined verdict's reason, the fact gate's block and the source list change, the send-back for tone, the approval, the pull request and its merge, the detection and the lead's note. The Performance tab's range for the cycle. At the monthly review the lead will read the range against the baseline sample and the previous cycle, decide next month's cluster from the verdict log, and promote nothing on publishing.
The library's piece on how a Search Visibility FTE works describes each mechanism outside the story; the worth-it verdict framework describes the four questions the two verdicts answered.