Pinned Tweet
One question, 2–6 AI models. Each answers alone, then they critique each other anonymously. One synthesis shows where they agree and where they don't.
That is Consensus, one of six modes in ewpire, with chat, code, images, video and music.
One credit balance, no subscription.
Ask your own question: ewpire.com/?utm_source=x&utm…
The account is free. The first run needs a top-up, from $10.
One automated code review, read out of the prompt that runs it rather than a report it produced.
It asks five things: does the build do what the brief said, is it correct, is it secure, would it build and run at all, and was it designed on purpose.
The bar is a confidence of 75 and no blocking flag standing. The flag needs a strict majority of the models that answered, and an exact split does not block, because there is no majority to report.
Below two answers, an older keyword scan decides instead.
Two models can only agree or disagree. There is no majority there, nothing to resolve a split.
So the rule under our idea checks is a floor, not a preference: at least three seats, and never all of them from one lab.
We generate your llms.txt and call it a hedge.
Ten minutes of plain text you publish yourself: what the product is, who it is for, what it costs. Cheap to put up, unproven as a channel, and the report prints that word beside the file rather than a forecast.
The file comes out of the presence run in Traffic, 5 credits, beside the answer-first copy, the FAQ and the JSON-LD. An audit reads how models describe you today and changes nothing published. Probability, never a promise. 20 free credits, no card: ewpire.com/traffic?utm_sourc…
In April 2025 OpenAI withdrew a GPT-4o update and wrote that it had produced "responses that were overly supportive but disingenuous".
The lab saw that from outside. Inside one chat, a yes that could never have been a no still reads like judgement.
Quote is OpenAI's own, on the update it rolled back, April 2025: simonwillison.net/2025/Apr/3…
Ideation runs several models on the idea instead of one – each alone, then reading the others with the lab names stripped. 20 free credits: ewpire.com/ideation?utm_sour…
The same question about your company, put to two assistants, and the two answers read side by side.
Both might draw a blank. Both might match, and a match is not proof – two models can be confidently wrong in the same way. Or they disagree, and that gap is the part to read.
This one is Traffic: several independent models get the same questions about your brand, and you see who knew you, who did not, and where they disagree. Never a promise, only a higher probability. From $25/mo yearly, 20 free credits, no card: ewpire.com/traffic?utm_sourc…
A single reviewer cannot disagree with itself.
The four builders we track each document one automated review. We put several models on the same file instead: separately first, then each reading the others with the lab names stripped, and the tally is shown.
$199 came from a version of our own page we replaced in July.
On 3 August the index was asked what our company costs and answered from a cached copy. No chat assistant was queried: they need a login. One engine refused outright. Web index and live pages only.
That run was a manual index check, not a Traffic audit. Traffic puts buyer questions about your brand to several frontier models and shows where they contradict each other. It raises the probability of an accurate description; no citation is promised. ewpire.com/traffic?utm_sourc…
Being unknown is the cheaper problem, and publishing is what moves it.
A description that is confident and wrong is the expensive one. It gets copied into the next one, it reaches you last, and nobody tells you it is out there.











