Step 4 of 4 · Stages 05–06
Publish & Observe
Closes the loop
Re-observe feeds directly back into Stage 02 — Evaluate. This is what closes the loop: the same questions get asked again against your updated sources.
Publish the signed-off draft to your configured destinations, then re-run the same prompts on the same models and compare. Treat a move as real only once it is larger than the variation you get re-running those prompts unchanged.
- Publish the approved draftout to your configured destinations
- See where you can publishyour own site, Senso, or both
- Re-read the numbers latersame prompts, after the re-crawl
- Compare to your baselineagainst step 2, not against zero
Produces a published page, and a new reading.
How to run it
| Step | API call | Why it matters |
|---|---|---|
| Publish the approved draft | POST/org | Ships the approved draft to your configured destinations. It needs the id of a tracked question, which step 2 creates, so a brand-new organization has to add one before this call succeeds. |
| See where you can publish | GET/org | Lists the destinations you can publish to and how much is live at each: your own site, a Senso agent-readable destination, or both. |
| Re-read the numbers later | GET/org | Returns mentions, citations and share over a window you choose. Give models time to re-crawl first — a same-day reading measures the crawl, not the change you made. |
| Compare to your baseline | GET/org | Returns the same per-question numbers step 2 recorded, so you can difference them. Treat a move as real only once it is larger than the variation you get re-running the same prompts unchanged. |
Stage 05 — Publish
- 1Choose the destination.Your own site, a Senso agent-readable destination, or both.
- 2Approve the publish.Explicitly, every time. Not inferred from the draft being approved.Requires a human
- 3Keep the receipt.It is how you prove later what was checked and who signed off.
Receipt
Portable proof the check happened. It records what was examined, what the evidence was, who approved it, and where it went.
Full definition →A receipt records:
- Receipt id, organization, and what it is about
- Content version, and every claim that was checked
- The evidence behind each claim, and its result
- Who reviewed it, and when it was verified and published
- Destination, published URL, and whether provenance is complete
The one prerequisite
Steps do not otherwise wait on each other, but this call does: publishing needs the id of a tracked question, and a fresh organization has none. Add one in step 2 first.
Stage 06 — Re-observe
- 1Re-run the same questions on a schedule.Same questions, same models. Changing them makes the comparison meaningless.
- 2Compare against your baseline from step 2.Citation lift, answer quality, and whether agents are crawling what you shipped.
- 3Record model answers as observations.Store them separately from your approved sources. They are evidence, not truth.
- 4Turn what you find into the next set of gaps.Feed them back into step 2. That is what makes this a loop.
- 5Check context before a consequential action.Price, policy, product details, eligibility, terms, and who is authorized. Then attach the receipt.
What you measure
Mention rate
Whether you were named at all.
Citation Rate
Of answers that cited anything, how many cited your tier.
Citation Share
Of all citation instances, what fraction was yours.
Share of voice
Your presence next to competitors.
Average rank
Where you appeared in the answer.
Crawl activity
Whether agents are fetching what you published.
Being mentioned and being cited are different, on different denominators. Improving one does not automatically improve the other. Analytics has the exact math.
What is not promised
That models will cite something new within any fixed period. The claim is narrower: what a model says about you can be traced to something you published, and you can measure whether publishing moved it.
The rule that protects everything upstream
Observation vs ground truth
Your approved records are the source of truth. Model output is an observation. An observation decides what to work on next; it never overwrites a source.
Full definition →A model answer that contradicts your context layer means something needs attention, not that your records are wrong. Write observations into your sources and the context layer degrades into whatever the open web already believed.
Before an action completes
Same context layer, different job. Before a booking, quote, application, product selection or checkout, verify what supports it, then attach the receipt so you can prove afterward what the decision was based on.
How to reach it
/org/content-engine/publishOpen in API reference →Publish verified content to your configured destinations./org/destinationsOpen in API reference →List where you can publish, with live counts per destination./org/analytics/summaryOpen in API reference →The rollup: mentions, citations and share, over a window you choose./org/analytics/citations/domainsOpen in API reference →Citation Coverage per domain. Whether your own pages are the ones cited./org/analytics/citations/pagesOpen in API reference →Down to the individual page, so you can see if what you shipped is working.Done when
- There is a live URL, approved by a person who is recorded as having approved it
- The receipt is retrievable
- The same questions run on a schedule, so change is measurable
- Model answers are stored as observations, separate from approved records
- The cycle produced the next set of gaps, not just a report