Generate the Traffic Quality PDF report

The PDF report is the evidence you send a publisher: page one is the ad server’s own verdict on the placement-day, exactly as the Details panel shows it, and AI-written analyst notes follow it with a neutral message for the publisher. A report is made once per placement and day and kept, so a colleague who downloads it later gets the same file. The verdict itself and how to read the page are on Read the Traffic Quality report.

Example. The Sports Daily › sidebar › 300×250 placement reads Act on yesterday’s scan with 34 % invalid traffic and $41.20 of its $132.80 gross paid for invalid impressions. You click Generate report on the row; when the report is ready the page offers Download PDF, and the analyst notes end with a message you paste into your email to the publisher.

Typical scenarios

You want	Do	Where
Get something you can send the publisher	Generate report on the row, then Download PDF: page one is the verdict, the analyst notes end with a neutral message for the publisher	The Report column
Add the analyst notes to a report made while no key was at hand	Add AI notes on the row, once there is a key to run them on and allowance left this month	The Report column
Let an AI assistant draw its own conclusions without paying for the notes	Run the deep analysis on its own over the API, without the AI findings and the PDF	Over the API

Step by step

1. For evidence, click Generate report on the row. The button becomes Generating…; the page polls every 30 seconds and, when the report is ready, asks Report for placement N is ready. Download now? and then shows Download PDF on the row. Rows already generating when you open the page keep polling. The report is made once per placement and day and kept: downloading it again, or anyone else downloading it, reuses it. A report made without the analyst notes shows Add AI notes beside Download PDF once there is a key to run them on and your monthly AI allowance is not spent.

The PDF report. Page one is the ad server’s own verdict, as the Details panel shows it: the status, the invalid share with its likely range and split, the facts, why, what to do and the context. Analyst notes written by AI follow it and never replace it: a headline, whether the AI agrees with the status — it may say the status may overstate or may understate the problem, and why — its findings, up to three recommendations and a neutral message for the publisher you can paste into an email. The raw data is an appendix on a page of its own; once the deep analysis has run it also lists the day’s refused and flagged events by reason, with refusals logged without a reason as Reason not recorded. The report is written in the language of the user who asked for it when the PDF can set that script — English, German, Spanish, French, Portuguese, Indonesian and Ukrainian; a user on Arabic, Hebrew, Urdu, Hindi, Bengali, Chinese, Japanese or Korean gets it in English. A stored report keeps the language it was written in: a colleague on another language downloads the same PDF rather than paying for a second write-up.

Who pays for the analyst notes. They are one model call per report, made on the key of whoever asks for the report, the way the AI assistant picks it: their own Anthropic key, else the network’s, else the platform’s while the network’s billing plan switches it on — then billed to the network at the list price × the plan’s AI price multiplier, and listed as Traffic Quality report in the AI usage report on the Ad Server Fees page. They count against the monthly AI allowance of whoever asks for the report, the same one their AI assistant counts against. With none of these keys, or with that allowance spent this month, the report is still made, without the analyst notes and without a model call, and a note above the grid says why (You have reached your monthly AI assistant budget for the allowance); the notes can be added to it later with Add AI notes, once there is a key and allowance left. Until release 155 the notes always ran on Epom’s key, free of charge.

Caveats and exceptions

• A report without notes stays requestable. A report made while no key was at hand — or whose AI call failed — has no analyst notes. Once a key is available to whoever asks, Add AI notes on the row (or the API and the assistant’s traffic_quality_analyze with report: true) writes them: only the notes are a new, billed model call; the stored deep analysis and figures are reused, and the PDF is rewritten with the notes. Without a key the row offers only Download PDF. A report that has its notes is final and is never paid for again.
• PDF generation gives up silently after about 20 minutes (40 polls, 30 seconds apart) with no error. If a row never turns ready, generate it again rather than waiting. Only one report is built at a time network-wide.
• A row with a report keeps its numbers. A day that is scored again refreshes its rows, except rows that already have a PDF report: they keep the figures the report was built from.

The caveats that apply to every Traffic Quality figure are on Read the Traffic Quality report.

Not working?

The full list of causes is on Read the Traffic Quality report; the ones about the report itself:

Symptom	Check	Where
Generate report never turns into Download PDF	Generation gives up silently after about 20 minutes; only one report is built at a time. Generate it again.	The row
The report has no analyst notes, and a note above the grid says why	No key (own, network, or platform with the plan’s AI switched on), or your monthly AI allowance is spent. Add AI notes appears once a key and allowance are available.	AI assistant keys; your monthly limit

Reference

The grants and the key the notes need are in Permissions; the REST endpoints and the assistant’s traffic_quality_analyze tool, which also writes the notes onto a report that has none, are under Over the API.

Related

• Read the Traffic Quality report
• Read the Traffic Quality cards, rows and Details
• AI assistant — which key the analyst notes run on
• Billing, ad-server fees and referral billing