How a candidate found out what AI says about them
Voters now ask assistants where candidates stand. One campaign used Seeqer to see the answer - and learned the difference between what they are and what AI thinks they are.
A composite user story. The person, the office and the country are deliberately unnamed; the pattern is what matters.
"Just ask it where I stand on energy"
The request came from a campaign manager, three months before a regional election. Their candidate had been told by a young volunteer that half her friends no longer read manifestos - they type a question into an assistant and read the paragraph that comes back. Something like: "Where do the candidates in my region stand on energy prices?"
So the team asked. Four engines, one question. The answers were fluent, confident, well organised - and, for their candidate, wrong in a way that was hard to argue with. Two engines summarised her as primarily a housing politician. One led with a controversy from an earlier term. None of them mentioned the energy programme that had been the core of every speech she had given that spring.
Nobody had lied. The engines had simply read what was readable.
The distinction that made the project possible
What an AI engine returns is not a portrait of a person. It is a summary of the material about that person that the engine can reach and trust: their own website, party pages, interviews, archives, news coverage. It describes perception, not conviction. Confusing the two is a mistake - but ignoring it is a bigger one, because for a growing share of undecided voters, that paragraph is the first and sometimes only description they read.
Four versions of the same person
The gap was not between truth and falsehood. It was between four sources that each told a partial story.
The candidate
Ran on four themes, with energy costs as the centrepiece of every speech that spring.
The website
Led with a housing programme published two years earlier, still the most linked page on the domain.
The coverage
Repeated one old controversy in almost every profile piece, because that is what archives are full of.
The AI answer
Blended all of it into three tidy sentences that felt authoritative, and that no one on the team had written.
Perceived emphasis vs. intended emphasis
The team ran roughly forty voter-style prompts across four engines, every day, and recorded which topic each answer attached to the candidate. Then they put that next to how the campaign itself weighted the same topics.
Illustrative example, not measured data. The shape of the gap is the finding, not the exact numbers.
What they did with it
- 1
They fixed the source, not the answer
The position papers on energy existed - as PDFs behind a menu, with no plain-language summary. Rewritten as readable pages with dates, numbers and a clear stance, they became something an engine could quote.
- 2
They gave the old story a current counterpart
You cannot delete an archive. But when recent, substantive material on the same topic exists, engines have something newer to cite - and the balance in the answer shifts.
- 3
They found the real demarcation lines
Running the same prompts for the other candidates showed which topics were already crowded and which were genuinely unclaimed. Two of the four planned contrast themes turned out to be indistinguishable in AI answers. The team dropped them.
- 4
They watched it move
Daily runs over the following weeks turned the work into a curve instead of a guess: energy mentions rising, the controversy fading from lead sentence to footnote.
The same pattern shows up far outside politics. See where teams use it.
What the campaign manager took away
Her summary, at the end of the project, was blunt: "We spent years managing what journalists write about us. We had never once looked at what the thing answering our voters' questions writes about us."
None of this changed who the candidate was. It changed which parts of her record were reachable, current and clearly stated - and therefore which parts an engine could repeat. That is a narrow, honest kind of influence: you cannot argue an AI into liking you, but you can make sure the accurate version of your position is the easiest one to find.
Whether you are working against an established public perception, sharpening a position, or looking for the demarcation lines that genuinely separate you from the competition, the first step is the same - read the answer people are actually getting.
See how AI describes you
Enter a website address. Seeqer builds the prompt set, runs it across the major AI engines, and shows you what comes back - daily, and next to whoever you are compared with.