Does the AI answer match your facts?

Requires an organization and a connected agent — Start Here covers both. Steps 1 to 4 run from the CLI; step 5 is read from the gap report.

1. 1Import your websiteThe agent infers your industry, pulls the relevant pages into your knowledge base, and returns the questions that industry is tracked on. Set market and language — the same question has a different answer in each.
2. 2Select a tracked questionChoose one where a competitor is cited and you are not.
3. 3Run the evaluationOne model call, roughly thirty seconds. Batches run in parallel, so fifty take about as long as one.
4. 4Add the missing sourcePull the page from your own site, upload a document, write the fact, or approve the model’s answer. The previous version is kept.
5. 5Check factual accuracyThe question stops counting against the score once the gap is closed.
6. 6Generate a draft and re-checkWritten from your sources and evaluated against them.

The loop above is three agent skills and a handful of commands. Install the skills to let your agent run them, or run them yourself.

# The three skills this loop uses
senso skills install search ingest content-gen

# Or take one at a time from Shipables
npx shipables install senso-ai/senso-search --claude

# 2 — the questions monitored across models
senso questions

# 3 — what your own sources say about one of them
senso search "Can a family of four stay together, and what beds are included?"

# 4 — add the missing source
senso kb upload ./family-suite.md --folder-id <folder-id>

# 6 — draft from the corrected source, then see what is waiting for review
senso generate run
senso content verification

Every command is listed on the CLI reference, grouped by the playbook step it belongs to. Step 5, the score, has no CLI command yet — it is computed on demand and read from the gap report.

Every tracked question comes back with three facts.

Four routes, cheapest first. The finding tells you which one applies.

1. 1Pull the page you haveFastest fix, and the right one when a page exists but was not cited.
2. 2Upload the documentFor a fact written down internally and never published.
3. 3Write the factInclude what changes it: dates, minimums, exclusions, what is sold separately.
4. 4Approve the model’s answerOnly where someone who owns that fact confirms it. The earlier version is kept.Requires a human

Of the questions asked of your knowledge base, the share it could answer without a finding against it.

• •The denominator is questions asked of your knowledge base — by a person searching, by your API, or by an agent drafting.
• •Computed on demand from the database. Nothing is stored and no nightly job recalculates it.
• •The check finds about 92% of the issues present in testing. When it is wrong it misses one rather than inventing one.
• •There is no brand alignment score. A brand check can be called, but nothing records its result.

It is not a citation metric. Citation Rate, Coverage and Share ask what got cited and divide by different denominators — see the citation metrics glossary.

1. 1The question“Can a family of four stay together, and what beds are included?” Kizuna is not mentioned; two other sites are cited instead.
2. 2What the knowledge base holds“The Family Suite accommodates four guests.”
3. 3The findingCapacity is supported. Bed configuration has no source — a gap, not a falsehood.
4. 4The contribution“Four guests in two queen beds. Guests share the two beds; four separate beds are not provided.” Approved by the owner.
5. 5The score movesThe gap is closed, so the question stops counting against you.
6. 6The draftComes back with the bed configuration in it, checked against the source that now exists.

• •Not cited is not wrong. It means your pages were not the ones used.
• •No source is not false. It means you hold nothing that settles it either way.
• •Saved is not verified. A model sentence becomes a fact when someone accountable says so.
• •A higher score is not a changed public answer. Publishing and being picked up takes weeks, like SEO.
• •Connecting an agent platform lets you use Senso from inside it. It does not change what that platform tells other people, and every user installs the connector themselves.
• •Publishing to the web is gated until an organization is verified.