Read a Buyer questions report

A collection run produces a Buyer questions report. The free report creates a collection called Buyer questions and runs it once, which gives you this report. See Get your first AI report. Every collection you build gives the same kind of report, and calls each of its journeys a buyer question.

What to look at first

1. The headline. How many buyer questions you win, and how many could not be judged. If several could not be judged, check that Company Profile has your brand names before you read further.
2. The You lose group at the top of Every buyer question. Open a row and read An answer that did not pass.
3. Recommended instead of you: who assistants put forward when you did not win.
4. Fix these first: up to three changes to start with.

How each result is decided is on How results are measured.

The Buyer questions report, in order

While the run is still going, the page says This run is still collecting answers. The report appears when every question’s answers are in, or after 60 minutes pass without a new answer.

Part	What it shows
Title and line	“Do AI assistants recommend Acme?”, then the number of questions, how many times each was asked, on which assistants, and the date. With more than one run, a pager reads “report 2 of 3”.
Share and Download PDF	Share creates a link anyone can open without signing in. Download PDF opens your browser’s print dialog and creates a share link if the report has none. See Collections and sharing.
Answers collected	Shown only when the run stopped early: how many answers were collected. The counts use those answers.
Headline	“Assistants recommend Acme in 14 of 15 buyer questions”, a square per question (filled for You win, half filled for Split, an outline for You lose, dashed for No result), and a written summary.
Recommended instead of you	In the answers you did not win: how many named each competitor in your Company Profile. Under Also named, not on your list, other products those answers put forward, each with Add to competitors. With no competitors in Company Profile, the report asks you to add them.
Fix these first	Up to three next steps. Each says how many buyer questions it affects and has Start the fix, which starts a Docs Agent or Marketing Agent session.
Every buyer question	A row per question, grouped You lose, Split, You win, then No result. Columns: Buyer question, answers passed per assistant, Passed (for example “31 of 38”), and Recommended instead. The last row, Answers passed, totals each column.
Where assistants get answers	A pie chart and a table of the sites cited, by Kind of site: You, Competitors, Review sites, Community, Blogs and news, Other sites. You and Competitors are split into docs and website. A page counts once per answer.
How we tested	Closed by default. The assistants and their models, the number of questions and answers, web search, the date, your sites, your competitors, and How results are decided.

An open buyer question

Click a row in Every buyer question to open it:

• Asked as: the persona the agents played.
• The conversation: each turn of the question, with how many answers named you at that turn, for example “Named in 12 of 40”.
• Checks, worst first: each check, with how many graded answers passed it, for example “Passed in 31 of 38”.
• What went wrong: the causes of failed checks, each with Fix in docs or Draft in Marketing.
• An answer that did not pass, with Read the full answer.
• Open the journey report: the full report for this question.

On a later run, a row can carry New, Question changed, Was a win, Was split, Was a loss or Had no result, compared with the collection’s first run.

A single journey’s report

Open the journey report in a row, or Open on a journey’s Reports page, shows one journey’s report. It has no win or loss; it shows each check on each run.

Part	What it shows
Header	The title and question, the number of runs, the date, and how many errored. baseline marks the first report for that agent and model.
Showing	With two or more agents: All engines, or one agent at a time. Agents are never averaged together.
Scorecard	A row per check, a square per run: green passed, red failed, a white outline not checked, grey errored. Each row shows how many runs passed. Click a square to read that run.
What went wrong	Causes ranked by how many runs each explains, each with evidence quotes and a recommendation, and Fix in docs or Draft in Marketing.
Who was named	Shown when the journey has a brand or competitor check: you and each competitor, with how many runs named them.
Where it read	Shown when an answer cited something: the domains cited, with the share that are yours. Each domain opens to its pages.
Facts	Totals across the runs: turns, sources cited, the share that are yours, and tokens.
Run configuration	Closed by default: the agents and models, the number of runs, web search, the date, and your docs sites.
Trend	Other reports for the same agent and model.

A run’s page shows the agent, model and tokens, each check’s result with the judge’s reason, and the conversation with the searches the agent ran and the sites it cited. On a journey with several turns, some agents record only the final answer; the run page then says that replies to earlier turns are not shown.

FAQ

Why does the report say “buyer question” when the app says “journey”?

They are the same thing. The report is written for readers who have not used the app, including people you share it with.

Why are some sources not marked as mine?

The report recognizes your pages from the documentation sites in Company Profile. Add each docs site there. Codex reports no sources, so its answers never appear in the sources sections.

Can I trust the written summary and the fixes?

An AI writes them from the report’s numbers, and the report says to check them before acting on them. The numbers themselves come from the checks.

Who can see a report?

Members of your organization. Anyone else needs a share link. See Collections and sharing.