free scan · public files only · result by email
Enter a domain. We fetch ads.txt and app-ads.txt, pull sellers.json for every seller you list, read /.well-known/adagents.json, check robots.txt and llms.txt for the AI crawlers, and detect Prebid.js on your homepage. No login, no upload, nothing from your ad server.
GET /ads.txt200212 lines
1google.com, pub-0000000000000000, DIRECT, f08c47fec0942fa0 # DIRECT · sellers.json match · PUBLISHER2ssp-one.example, 11822, DIRECT # DIRECT · sellers.json match · PUBLISHER3ssp-two.example, 48213, DIRECT, 0000000000000000 # DIRECT · sellers.json match · BOTH47relay-ads.example, 7781, RESELLER # relay-ads.example serves no sellers.json48relay-ads.example, 7781, RESELLER # duplicate of line 4791openhop.example, 10442, RESELLER, 0000000000000000 # seller_type=INTERMEDIARY · is_confidential=1130ssp-two.example 48213 DIRECT # malformed · missing separators212# end of fileAn agent reading example.com today finds no adagents.json.
Everything below is fetched over HTTPS from your domain and from the seller domains your ads.txt names. Nothing comes from you.
| File | What we check | A flag looks like |
|---|---|---|
| /ads.txt and /app-ads.txt | Line count. DIRECT versus RESELLER count. Duplicate lines. Malformed lines. RESELLER entries with no matching sellers.json record. Seller domains that return no sellers.json at all. seller_type recorded as PUBLISHER, INTERMEDIARY or BOTH. Entries whose seller is marked is_confidential. | 9 RESELLER lines point at 4 seller domains that serve no sellers.json |
| sellers.json, one per seller domain in your ads.txt | Fetched for every seller domain you list. Each of your lines matched back by seller_id. Type and confidentiality recorded per entry. | seller_id 10442 at openhop.example is INTERMEDIARY with is_confidential=1 |
| /.well-known/adagents.json | Present or absent. Valid JSON against the AdCP schema; a parse failure is a finding. Number of authorized agents. Agents with no reachable endpoint. Whether your own domain appears as an authorized seller agent. | adagents.json returns 404 |
| /robots.txt and /llms.txt | Rules for GPTBot, ClaudeBot, PerplexityBot, Google-Extended and the rest of a crawler list kept in a config file because it changes monthly. llms.txt present or absent. | Google-Extended disallowed on / · llms.txt absent |
| Prebid.js | Detected on the homepage or not, from one headless fetch. Version string. Number of bidder adapters configured, where pbjs.getConfig or adUnits is readable from page source. | Prebid 9.14.0 detected · 14 bidder adapters |
Every raw check result is stored, not just the score. If a file is missing, times out, or fails to parse, that check records its own failure and the scan continues.
Every check writes its own result, pass or fail, with the detail that produced it. The three scores are computed from those results and weighted equally for now. Because every raw result is stored, the weights can change later without crawling you again.
Moves with duplicate and malformed lines, RESELLER entries with no matching sellers.json record, seller domains that publish no sellers.json at all, and sellers marked is_confidential.
{ "seller_id": "10442", "seller_type": "INTERMEDIARY", "is_confidential": 1}Moves with whether /.well-known/adagents.json exists, whether it parses against the AdCP schema, how many of the listed agents have a reachable endpoint, and whether your own domain is listed as an authorized seller agent. If the file is absent, every check in this group fails, which is why the sample reads 0. Prebid.js detection is recorded alongside it, with the version and the bidder adapter count.
{ "authorized_agents": [ ]}A 404 and an empty array score the same. An agent finds no authorized seller agent either way.
Moves with what robots.txt says to GPTBot, ClaudeBot, PerplexityBot, Google-Extended and the rest of the crawler list, and with whether llms.txt exists. The score counts which crawlers can read you. Whether to open or close the gate depends on your contracts and your traffic. The report says what the gate says today, so the choice is yours and it is deliberate.
User-agent: GPTBotAllow: /
User-agent: ClaudeBotAllow: /
User-agent: PerplexityBotAllow: /
User-agent: Google-ExtendedDisallow: /Allow lines carry the good tint and Disallow lines the critical tint. The same three colors mark the score bands: critical under 40, warn 40 to 69, good 70 and up.
Subject: Scan result for example.com
Supply-path hygiene 61 · Agent readiness 0 · AI discoverability 74
Want the full revenue audit? Reply to this email.
One HTML email. No PDF, no portal, no login. The figures are the sample above.
The scan reads what is public. The audit reads your ad server. You send one Google Ad Manager export and your Prebid config. We send back a fixed-format report and walk through it with you on a call. Every number in it carries the assumption that produced it, as a low and a high.
Fill rate by ad unit, device and month, with every unit under your own median flagged. Unfilled impressions multiplied by your median eCPM is the unrealized-revenue estimate, and that assumption is printed next to the number. Your fill rate, eCPM by device and reseller count are set against benchmarks being built from weekly scans of 200 publisher domains and from prior audits. Where a benchmark comes from public sources instead, the report says so on the page where it is used.
Revenue share and eCPM for open auction, private auction, preferred deal, programmatic guaranteed and header bidding, month by month. Any channel whose share moved more than 10 points in six months is called out with the months it moved.
The scan's ads.txt findings crossed with your revenue by bidder and SSP. Resellers earning under 0.5% of revenue are listed as removal candidates. SSPs in your Prebid config but absent from ads.txt, and the reverse, are listed with the revenue at stake.
Timeout against norms, flagged under 1000 ms or over 3000 ms. Bidders configured with no revenue in the last 90 days. Price floors present or absent. User ID modules are listed for your review. Enabling them is a judgment call and the report says so.
Everything the scan checks, plus whether your Prebid setup includes the seller-agent module and whether it is configured, whether an agent in your adagents.json answers an AdCP discovery call, and whether your inventory and first-party signals are described anywhere a machine can read them. Our discovery client is read-only, and a test in the codebase proves it cannot start a transaction. The output is a gap list in the form 'to sell to an agent, you need X, Y and Z', each gap mapped to an action with a rough effort estimate.
Which of your inputs lead daily revenue and eCPM, at what lag from 0 to 60 days, and how strongly. Candidates are ad requests, fill rate, active bidder count, Prebid timeout changes treated as step events, and page views and direct-traffic share if you share analytics. Series are differenced or detrended first. A relationship has to hold in both halves of your date range. One that fails that test is not reported, and the table says it was dropped instead of leaving it out silently. Step changes in revenue or eCPM are listed with whatever else changed in the same window.
| leading signal | lag (days) | strength | holds in both halves? | plausible mechanism |
|---|---|---|---|---|
| ad requests | 9 | strong | yes | more requests reach auction before eCPM adjusts |
Your ad requests lead revenue by about 9 days, consistently. Your reseller changes do not show a measurable effect.
Every quantified prediction in the report is recorded at delivery. Sixty days later we score it against a fresh export as a hit, a miss or inconclusive, with the actual number beside the range.
Three rules hold in every report.
report structure, in bound order
Delivered as a PDF to a private Drive folder, with a readout call. No dashboard, no login, nothing to maintain after we leave.
Both tiers start with a short intake and a one-page sheet that tells your ad-ops team exactly which Google Ad Manager report to export. You drop the files in a private Drive folder. Nobody from your team logs in to anything.
GAM report
Prebid config
Gaps larger than two days and currency mismatches are flagged on ingest, before any analysis runs. Raw files stay in private object storage next to the normalized rows, so every number in the report can be reproduced. Google Ad Manager exports only, for now.
A finding without a number does not go in the report. Each one shows the evidence that produced it, an estimated revenue impact as a low and a high, the effort to fix, and a priority. We do not publish point estimates, because we would be guessing the second digit.
format example, not a result
Removing the six resellers below 0.5% of revenue should move eCPM by +3% to +7% within 60 days.
| Finding | Evidence | Impact (low / high) | Effort | Priority |
|---|---|---|---|---|
| Remove 6 resellers under 0.5% of revenue | 6 of 41 seller domains, 0.3% of 13-month revenue combined; 4 of the 6 are among the 31 unverifiable lines | eCPM +3% / +7% within 60 days | low | 1 |
| Raise Prebid timeout from 800 ms | timeout 800 ms, under the 1000 ms threshold; 3 of 14 configured bidders show no revenue in 90 days | revenue +1% / +4% | low | 2 |
| Fill under site median on 3 mobile units | 3 units at 71 to 78% fill against an 86% site median; 4.1M unfilled impressions × $0.92 median eCPM | $2.9k / $4.6k per month unrealized | medium | 3 |
At delivery, every quantified prediction is written down with its metric, its low and high, and a 60-day window. That record cannot be edited afterward. After 60 days you send a fresh export, and each prediction is scored with the actual number beside the range. Hit means the actual landed inside the range. Miss means it landed outside. Inconclusive means the follow-up data did not cover the window. All three stay on the list. A scorecard that only shows hits is marketing.
format example, not a result
There is no hit rate on this page because none has been earned yet. When five or more audits have been scored, the aggregate goes here, misses included.
predictions recorded 0 · scored 0 · hit 0 · miss 0 · inconclusive 0
Fixed fee per tier, quoted on the call. Deep is booked with a 50% deposit.
| What each tier includes | Rapid | Deep |
|---|---|---|
| Inputs | One GAM export and your Prebid config | The same, plus a fresh export at day 60 |
| Supply-path hygiene, crossed with revenue by bidder and SSP | included | included |
| Fill and match, demand channel mix | included | included |
| Prebid configuration review | included | included |
| Agent readiness at scan depth, AI discoverability | included | included |
| Fixed-format PDF report with the appendix, and a readout call | included | included |
| Agent readiness gap list with the read-only AdCP discovery check | not included | included |
| Leading indicators and step changes | not included | included |
| Predictions recorded at delivery and the 60-day calibration scorecard | not included | included |
Not sure which? Book the call and bring your scan result.
ads.txt, app-ads.txt, adagents.json, robots.txt or a Prebid config. The report recommends, in order, with effort and impact. Your team decides.Every report's appendix repeats this list for that engagement, with the data coverage we actually had.
pbjs.getConfig() dump or the wrapper file. The intake sheet says exactly what to request.adagents.json to see whether it responds. It cannot start a transaction, and a test in the codebase proves that.Run the free scan first. If the findings are small, you are done, and it cost you an email address. If they point at money, book a call. We will tell you which tier fits and send the sheet that says what to export.