Kyanos · Paper IV · White Paper · August 2026 · Revision 2 · 11 August 2026

Somebody Else
Wrote Your Record

A brief for candidates and officeholders: where answer engines get what they say about you, which of those surfaces you may touch and which you are forbidden to, and why a challenger's problem is the exact opposite of an incumbent's.

Abstract. An advocacy organization publishes the evidence other people are judged by. A candidate is judged by evidence other people published. That difference decides what AI presence work actually consists of for a campaign, and it is why guidance written for organizations transfers badly. This brief covers where answer engines find material about a candidate, which of those surfaces a campaign may edit directly, which it is explicitly forbidden from editing, and why the challenger with no record and the incumbent with an unalterable one are solving opposite problems. It makes no claim that any of this changes vote share; that experiment has not been run.
↑ Contents 01 · Subject, Not Source

The record about you
is mostly not yours

The companion paper to this one argues that an advocacy organization is sitting on an asset: decades of scorecards, endorsements and research that answer engines can draw on, if the material is published in a form they can retrieve.1 That framing does not survive the move to a campaign, because a campaign is on the other side of the transaction.

When a voter asks an answer engine about your candidate, the material it assembles was, in the main, written by other people: a Ballotpedia entry, a Wikipedia article, local press, a scorecard from an organization that rated your record, and the opposition's framing where that has been picked up. Your own website is one source among those, and frequently not the most retrievable one.

"An organization asks whether AI can find what it published. A candidate asks what AI found, and who wrote it."

This is not a reason for despair and it is not a reason for panic. It is a reason to be precise about which surfaces you can act on, because the ones you can act on are fewer than a vendor will imply and more than most campaigns use.

↑ Contents 02 · Surfaces and Access

Four kinds of access,
and the difference matters

Every source an answer engine might use falls into one of a few categories, and the categories are not about credibility. They are about one question: can you change it, and how? Treating them as interchangeable is the most common and most expensive mistake.

Kind of surfaceExamplesWhat you may do
Controlled Your own site, your own social accounts Anything. You own it, you write it, it publishes when you say so.
Direct-edit third party Wikidata and other claimable profiles Edit it yourself, disclosing any paid relationship. It publishes immediately.
Request-and-review Wikipedia You are forbidden from editing it directly. Conflict-of-interest rules require you to propose a change and let an unconflicted editor decide.
Submission-gated Ballotpedia Nobody outside their staff edits it — but there is a defined channel, and it is worth using.
Uncontrolled Press coverage, aggregators, opposition material Nothing directly. You can only add to the record elsewhere.
Fact base Government and institutional records: votes, filings, registrations Nothing. It is already true and already public.
The Google Knowledge Panel is not on the editable list

The panel that appears beside a search result looks like a profile and is not one. It is a synthesis Google assembles from the sources above. A verified subject may claim it and suggest changes; nobody edits it directly.2 Campaigns routinely treat it as a page to be corrected, and then cannot understand why the correction does not hold. The way to change a synthesis is to change what it is synthesized from.

↑ Contents 03 · The Challenger

Absence is not
a blank slate

A first-time challenger often has almost nothing an answer engine can retrieve: no voting record, no Wikipedia article, a Ballotpedia stub, and a campaign site three weeks old. It is tempting to read that as a clean slate — nothing bad is being said.

An answer engine asked about you will not return nothing. It will answer from whatever exists, which may be a same-named person in another state, a single local news item, your opponent's characterization, or a general statement about the race. Absence does not produce silence. It produces an answer assembled from the least specific material available.

The work this implies is closer to publishing than to optimization. There is no archive to make legible; there is a record to bring into existence, in places designed to be retrieved:

  • A biography with dates, places and roles, on your own site, stated once and consistently. Not a narrative — a record.
  • Positions as commitments, with specifics. Named legislation where it exists, a number where one applies.
  • Your Ballotpedia submission, completed. Their candidate survey exists precisely so subjects can supply what staff summaries lack, and material a candidate supplies takes priority over a staff-written summary.3 It is free, it is structured, and most campaigns skip it.
  • A Wikidata entry, which you may create and edit directly with disclosure — unlike the Wikipedia article you may not touch.

None of this requires a vendor, and none of it requires guessing at how any platform ranks anything. It requires that the true record exist somewhere retrievable, which for a challenger it usually does not.

↑ Contents 04 · The Officeholder

A record you
cannot withdraw

An incumbent has the opposite problem. The votes, the committee assignments, the sponsored bills and the constituent service are in government databases, published, permanent, and retrievable whether or not you ever mention them. Nothing about that record can be edited, and it should not be.

The exposure is interpretation. A vote cast years ago carries no context in the record itself: not the amendment it was attached to, not the deal it was part of, not the version of the bill that failed the week before. An answer engine summarizing that vote will supply context from somewhere, and the somewhere is whatever was written about it — including by people who wanted it read a particular way.

What this makes worth doing

The only durable response is to be the most complete published account of your own record. Not a defence of every vote — a plain, dated, specific explanation of the ones that need one, on a surface you control, in language that names the bill and the date. You cannot remove the fact. You can be the source that explains it.

This is also why an incumbent's AI work is never finished. New votes are cast continuously, characterizations accumulate, and what an engine said in January is not what it says in October — which is a monitoring problem before it is a writing problem.

↑ Contents 05 · Rules

What you may do,
and who to ask

Two constraints shape what a campaign can safely do, and they are commonly conflated.

Conflict-of-interest rules on third-party surfaces are not law; they are the rules of the surfaces themselves, and they bite. Wikipedia forbids a subject from editing their own article and expects proposed changes to go through an unconflicted editor. A campaign that edits directly is likely to have the change reverted and may draw attention to the article it wanted improved.

Campaign finance rules are law, and they are narrower in scope than most people assume. The Federal Election Commission's jurisdiction runs to candidates for federal office only; a state legislative race is governed by that state's law instead.4 Where they do apply, the questions that arise around AI presence work are the ordinary ones — who paid for the work, at what value, and whether an outside organization's activity on a candidate's behalf is independent.

This brief does not give legal advice

Whether any particular arrangement between a campaign and an outside organization constitutes coordination is fact-specific and belongs with counsel. Two distinctions worth carrying into that conversation: an entity's ability to contribute to a candidate is a separate question from its ability to coordinate with one, and neither settles the other; and an organization publishing its own endorsement rationale on its own surfaces is engaged in its own speech, which is a different activity from working on a campaign's material.

↑ Contents 06 · Timing

The channel that cannot
be bought in October

Campaign money arrives late and goes where it can be spent fast. Advertising absorbs it well; the measured effects of television advertising decay within days of airing, which is why late saturation buys so little per dollar.5

This channel has the opposite shape. A record has to exist before it can be retrieved, third-party surfaces move on their own schedules, and a Wikipedia proposal waits on an editor who does not work for you. None of that compresses because a check cleared in September.

There is a second reason to start early that has nothing to do with production time. Until you have measured what engines currently say about your candidate, you cannot tell whether anything you did changed it. A campaign that begins in January has a baseline and can watch for drift. A campaign that begins in October has an opinion.

The first week's work

Ask four engines the ten questions a voter would actually ask about your candidate, and write down what comes back. Most campaigns have never done this, and it costs an afternoon. Whatever you find, you will at least be arguing about the same text.

What follows from the answers depends on which of the two problems you have. If the engines return little, you are publishing a record. If they return a characterization you did not write, you are adding context to one you cannot remove. Those are different jobs, and the first step in both is finding out which one you are looking at.

Notes
  1. See The Endorsement Economy (Paper III), which covers the organization side of the same relationship — scorecards, endorsements, and the legal-entity structure that shapes what a C3, C4 or PAC may publish. Its section on the three entity types is the shared frame for both papers.
  2. Kyanos treats the Google Knowledge Panel as a source signal rather than an editable profile, on the basis that it is a synthesis Google builds from other sources: a verified subject may claim it and suggest edits but cannot edit it directly. Recorded internally as decision.google_knowledge_panel_is_a_source_signal.
  3. Ballotpedia operates a candidate survey through which a subject may supply material directly; that submitted material takes promotional priority over a staff-written summary. No outside party may edit Ballotpedia entries otherwise.
  4. The Federal Election Commission regulates candidates for federal office. Candidates for state and local office are governed by state law, which varies and is not covered here. This distinction is recorded internally as decision.fec.jurisdiction_is_federal_only. Nothing in this brief is legal advice.
  5. Gerber, Gimpel, Green & Shaw, "How Large and Long-lasting Are the Persuasive Effects of Televised Campaign Ads? Results from a Randomized Field Experiment," American Political Science Review 105(1), 2011 — a randomized Texas experiment in which advertising effects decayed within roughly a week. Discussed further in Paper I.
About the Author
Ed Forman

Ed Forman is the founder of Raise Presence LLC, which built Kyanos to measure and improve how progressive candidates and causes are represented across AI platforms. About Raise Presence →

Note: AI helped me research and draft this document. I have read every word and it has been processed by my (human) brain — I've checked what AI drafted for me, made the edits I wanted, and I stand behind what it says.