Notes on the Record · No. 1
The cheapest thing in the world is a quotation
By Edwin Danso, Founder and Chief Executive, OTA Media
14 September 2026
The cheapest thing in the world is now a quotation.
Not a true one. Any of them. For the price of a few seconds of compute, anyone can produce a sentence a prime minister never said, in a voice indistinguishable from hers, dated to a day she was in the room. The technology to do it is not exotic, it is not expensive, and it is not going away. It is on the phone in your pocket.
What has not fallen in price, at all, is proof.
Establishing that a particular person said a particular thing, on a particular date, in a particular room, and being able to show the record it came from, costs roughly what it cost twenty years ago. It takes archives, and sourcing, and somebody willing to say “I cannot confirm that” when the pressure is to say something else. The two curves have been diverging for a decade and they are now far apart.
That gap is the story. Everything else follows from it.
The asymmetry, plainly
A fabricated quotation travels at the speed of a share button. A verified one travels at the speed of a person checking. The first is free and instant. The second requires that somebody, somewhere, has kept the record, indexed it, and is willing to be held to it.
This is not a new problem. Newsrooms have always been slower than rumour. What is new is the ratio. When fabrication required skill, effort and a studio, the asymmetry was tolerable, because the supply of convincing fakes was limited by who could make them. That constraint has gone. The supply is now effectively infinite and the verification capacity is roughly what it was.
You cannot fix that by making people more sceptical. Scepticism without a reference is just a second kind of noise: a world where nothing is believed is not better than one where everything is, and it is considerably easier to govern badly.
What closes the gap is not doubt. It is a record.
What we built, and what it is not
YouSaid is a search engine over the primary spoken record of public leaders. As at 14 September 2026 it holds 5,251,393 dated, verbatim lines, drawn from fifteen national legislatures and the United Nations record reaching back to 1946, with 17,321 named speakers, of whom 6,534 are verified against an independent directory.
Every line is reproduced as it was said, attributed to a named person, dated, and linked to the source it came from. Nothing is summarised. Nothing is generated. There is no model between you and the record, writing a plausible version of what someone probably meant.
It is worth being precise about what that is and is not.
It is not a truth machine. It does not tell you whether what a leader said was correct, honest or wise. Those are editorial and political judgements and they belong to readers, not to software.
It is not complete, and it says so. Fifteen legislatures is not every legislature. A record that reaches back to 1946 in one chamber and eighteen months in another is uneven, and we publish what we hold alongside what we serve so that the unevenness is visible rather than implied.
And it is not a database pretending to be neutral. Deciding which records to pursue, which sources meet the standard, what counts as an attribution and what does not, are editorial decisions. We make them, we write them down, and we can be argued with about them. A record that claims to have no judgement in it is simply one that has hidden where the judgement went.
The discipline that matters most
Here is the rule we will not trade away, and the one that costs us most often.
Where a speaker cannot be identified with certainty, we publish nothing.
Not a best guess. Not a probable match with a confidence score in small type. It is enforced in the code rather than in a policy document, because a policy is a promise and code is a constraint, and under deadline pressure only one of those holds.
The corollary is just as important and easier to get wrong. When a quotation is not found in our record, we say that it is not in our record. We do not say it was never said. Those are different statements and the difference is the whole profession.
A great deal of the archive of human speech was never written down, never recorded, or was lost. Our not holding something is a fact about us. It is not a fact about the world. A verification tool that cannot tell those apart is not a verification tool at all. It is a new and more confident way of being wrong, and there is no shortage of those.
This is why the most useful thing the system does is sometimes to refuse.
Why a record needs a publisher
There is a version of this that is just infrastructure: an index, an API, a thing other people build on. That version is easier to run and harder to trust.
Records need someone accountable for them. Someone who publishes the corrections rather than quietly amending the row. Someone who states the coverage rather than letting the absence of a result imply completeness. Someone who can be complained to, and who has to answer.
When we get an attribution wrong, and we have, the wrong line is suppressed with a reason code rather than deleted, and the correction is published. Nothing is removed from the record silently. That is not a technical decision. It is an editorial one, and it is the reason we think of what we are doing as publishing rather than hosting.
Verification is slower than accusation. It has always been slower, and it always will be. The most any of us can do is narrow the gap, publish what we hold, be honest about what we do not, and refuse to guess.
That is the whole of it.
Notes on the Record is a monthly column from OTA Media on the primary record of public speech: what is in it, what is missing from it, and what it costs to establish that anyone said anything at all. YouSaid.ai is operated by OTA Media Limited, registered in England and Wales, no. 16472914. ICO registration ZC121510. ISSN 3153-7596. Corrections: yousaid.ai/corrections. Complaints: yousaid.ai/complaints.