Methodology
How we measure, and what we refuse to claim
Most companies that grade your marketing are also paid to run it. We are not, and that is the only reason our numbers are worth anything to you. Here is what we capture, how we decide two readings can be compared, and where our measurement stops. Figures are as of September 24, 2026.
We do not run your ads
Radveo has never managed a client's advertising. The Google Ads code in our product can read an account and cannot write to it, by design. No campaign button, no budget slider, no bid change. If we ever add paid media, the June 17, 2026 ruling on file binds it: flat fee, never a percentage of spend, and you pay the platform directly.
A firm that earns more when your spend goes up cannot also be the firm that tells you whether your spend worked.
What we capture
Since July 26, 2026 we have asked four AI engines the questions your customers ask, daily, and kept the answer text. As of September 24, 2026:
- 15,106 captured answers, on 61 days of capture, across 365 distinct questions
- Perplexity 4,132, ChatGPT 3,973, Google Gemini 3,662, Grok 3,339
- A business was named in 686 of those answers and absent from 14,420
- 11,457 of the 15,106 carried the source URLs the engine said it used
When your report says an engine recommended someone else, you can read the sentence it said it in.
Where our measurement stops
Grok is in the roster and it is not grounded. Of its 3,339 answers, zero carried a source URL. Perplexity carried them on 100 percent, Gemini 98.2, ChatGPT 93.9. Grok tells us what it said, never what it read, so we count it, label it, and never call it evidence.
Google's local results are a cached snapshot. The map results on our city pages come from SerpApi, cached seven days. Those pages stamp a month and promise no refresh schedule, because we will not commit to a cadence we have not held.
Your free report is not indexed. On August 17, 2026 we blocked every crawler from every report URL. Of 515 stored verdicts, 474 said the business could not be seen, produced by a question that excluded it by definition, and 112 of the subjects were other people's businesses. Indexing them was our best growth channel. Your link still works for anyone you send it to.
What a percentage on your report means
Across everything captured, a business was named in 686 of 15,106 answers. The honest sentence keeps both numbers in it rather than saying 4.5 percent.
Your weekly report works the same way at a smaller scale. A line reads "named in 9 of the 12 answers we checked on Perplexity this week," never a bare 75 percent. At a denominator of twelve, one answer flipping moves the figure eight points.
How we decide two readings can be compared
Two readings are comparable only if both asked the same questions of the same engines. We fingerprint both, and if either differs the report says the sample changed instead of drawing a line between them.
One client has two readings from the same day: six percent and 75 percent. Nothing about the business changed. The questions did: nine broad ones in the first reading, twelve narrower neighborhood ones in the second, and a business gets named far more often in the narrow set. Charted together, that looks like our work. It was not.
The five rules every weekly report follows
- Every percentage carries its denominator.
- A gap is a named gap with a fix, never a zero for something we cannot see.
- One ask per email. More than one and none get done.
- A flat week says flat, and why that is normal.
- A down week leads the email.
Claims we refuse to make
On the record, so a hard quarter has to argue with a published page. The machine-readable version is at radveo.com/reality.txt.
We will never guarantee a ranking, an AI mention, or a result.
Nobody controls what an AI says.
We will never manage the advertising we grade you on.
Measurement paid by spend is not measurement.
We will never bill a percentage of ad spend.
It pays us to raise your budget.
We will never show a number we did not measure.
Real or absent, no third option.
We will never print a percentage without its count.
Alone it hides how small the sample was.
We will never compare two readings that asked different questions.
Method changes are not results.
We will never publish a report about a business that did not ask for one.
Ours said things about real companies that were not true.
We will never invent a source, statistic, review, or client.
A source nobody can open is a fabrication with a citation stapled on.
We will never publish content on your site without a person approving it.
Your name is on it.
We will never make a health, medical, or efficacy claim for a regulated client.
For cannabis that is WAC 314-55, enforced by a gate before a human reads the draft.
Take none of this on trust. Run the free check on your own business, read the answers the engines actually gave, and hold this page against what lands in your inbox.