The Correction Playbook: How to Fix a Wrong Fact About You in Google, Journalists' Archives and AI Answers

Start by finding where the error actually lives

When a founder spots a wrong fact, the instinct is to fix the place where they saw it. That is usually the wrong place. A wrong funding figure in a ChatGPT answer is rarely invented from nothing; it is typically a blend of what the model absorbed from published sources, plus whatever a search tool retrieved at the moment you asked. If you only patch the symptom, the cause keeps regenerating it.

So the first hour is diagnosis, not outreach. Open a spreadsheet and log every place the wrong fact appears. Check these, in this order:

• The original article or articles. Search the exact wrong phrase in quotes, plus your company name. Note the publication, author, date and URL.
• Secondary copies. Aggregators, syndicated versions, newsletter roundups and 'top startups' lists that repeated the original.
• Structured databases. Crunchbase, LinkedIn company page, Wikipedia or Wikidata if you have an entry, your own website's About and press pages, app store listings.
• Search results. The knowledge panel, featured snippets and 'people also ask' boxes.
• AI answers. Ask ChatGPT, Perplexity, Gemini and Claude the same three questions about you, in a fresh session each time, and screenshot the answers with the date.

The log matters for a practical reason: it tells you which source is upstream. If the wrong figure appears in one 2022 article and eleven pages that quote it, you have one real problem and eleven echoes. Fix the one and the eleven often matter far less.

Rule one: fix your own house before you email anyone

Before contacting a single journalist, make every source you control say the same correct thing. This is the cheapest step and the one most people skip, because it feels less urgent than confronting a publication.

Audit your website, your LinkedIn company page, your founder bio, your Crunchbase profile, conference speaker pages and any press kit. The goal is entity consistency: the same legal name, the same one-sentence description, the same founding year, the same title, the same funding figure, written the same way everywhere. Search engines and AI systems reconcile entities by comparing sources. If your own pages disagree with each other, a model has no reason to prefer the correct version over the wrong one.

Two concrete moves help here. First, publish one plain, factual page, such as an About or Company facts page, that states the key facts in full sentences, with dates. Second, add Organization and Person structured data to it so the facts are machine-readable. This does not guarantee anything, but it gives retrieval systems a clean, authoritative source to find. Budget two to four hours of work, and nothing beyond staff time.

Rule two: ask for a correction or an update, and know which one you are asking for

This is where most angry emails go wrong. A correction and an update are different requests, and journalists treat them very differently.

• A correction is for something that was wrong when it was published: a misquoted figure, a wrong title at the time, a misattributed claim. Newsrooms have a process for these and a professional obligation to take them seriously.
• An update is for something that was true then but is not now: you raised a new round, changed your CEO, rebranded, moved HQ. Most archives will not rewrite a dated article to reflect later events, and they are right not to. Asking them to treat it as an error gets you refused.

Do this, not that. Do not write 'your article is wrong and damaging our reputation.' Do write a short, factual note to the reporter first, and to the corrections address or standards editor if the reporter does not respond. A good request has four parts: the URL, the exact sentence that is wrong, what the accurate statement is, and a document that proves it, such as a press release, filing, or registry entry. Make it a thirty-second job for them to verify. Keep the tone neutral. Never threaten, and never demand removal when you are asking for a fix.

For true updates, ask for something smaller: a clarifying editor's note, or an added line such as 'The company has since raised additional funding.' Some outlets will do this; many will not. Either answer is fine, because the better route for updates is the next rule.

Rule three: outweigh the wrong source with newer, better ones

You will not get every outlet to change an archived article, and you do not need to. Search engines and AI systems weight freshness, authority and agreement across sources. A single stale article repeated by nobody is weak against several recent, consistent, credible ones.

That means the durable fix is fresh, accurate coverage. A founder profile, a byline article or a commentary quote in a respected outlet that states the correct facts gives retrieval systems something better to cite. This is slower and costs more than an email: realistic timelines for earned placements are weeks, not days, and fees or retainers vary widely depending on whether you use an agency or do it yourself. If you are choosing where to spend limited budget, a small number of accurate placements in outlets that AI tools tend to cite is a better use of money than trying to scrub old pages.

Rule four: handle the knowledge panel and structured sources separately

A Google knowledge panel is not edited by emailing Google. If you have a verified panel, you can claim it and suggest edits directly. If you do not, the panel is built from sources such as Wikipedia, Wikidata, official websites and trusted databases, so the fix is upstream: correct those sources, then use the 'suggest edits' feature once you have access.

Be careful with Wikipedia. If you have a page, do not edit it yourself where you have a conflict of interest. Use the article's talk page to propose the change with a reliable source cited, and disclose your connection. Self-editing is the quickest way to have a correct fact reverted and your motives questioned. Wikidata and Crunchbase usually have lighter processes, and both let you submit sourced corrections. Allow one to two hours per platform and expect days to weeks for review.

Rule five: for AI answers, feed the sources, do not argue with the chatbot

Telling an assistant 'you are wrong' inside a chat changes nothing for the next person who asks. Most assistants do not learn from individual conversations in a way that fixes the answer globally.

What works is addressing why the system produced the answer. Search-connected tools such as Perplexity, and ChatGPT when it browses, answer from retrieved pages, so they improve when the pages they retrieve improve. That is why rules one through four come first. Models that answer from training data update only when they are retrained, which you do not control and cannot schedule, so for those the realistic aim is to make sure the web is correct by the time the next version is built.

Most assistants also offer a feedback or report option on an answer. Use it, attach the correct fact and a source, and treat it as a low-cost extra, not a fix. It takes two minutes per answer and its effect is uncertain. Do not pay anyone who promises to 'delete' or 'overwrite' what an AI says about you. No one has that control.

Rule six: re-verify at four to eight weeks, then decide

Corrections do not propagate on demand. Newsrooms act in days to weeks, search indexes recrawl on their own schedule, and AI retrieval depends on what has been refreshed. Checking daily will only frustrate you.

Set a calendar reminder four weeks out and another at eight. Repeat the exact same questions you logged at the start, in fresh sessions, and compare against your dated screenshots. Record what changed. A realistic outcome is mixed: the original article corrected, the secondary copies lagging, one assistant fixed and another still wrong. Chase only the sources that are still wrong and still upstream.

When to let it go

Not every error deserves your time. Let it go when:

• The error is small, such as a slightly off employee count or a founding year that is a year out, and it does not affect a deal, a hire or a customer decision.
• It appears on a low-authority page that nothing else cites.
• The only way to fix it would be to start a public dispute that draws more attention than the mistake does.
• It is an out-of-date but once-true fact in an archived article. Publish the new fact instead.

Escalate, with a lawyer where appropriate, only when a statement is both false and materially harmful, for instance a wrong claim about legal trouble, insolvency or fraud. That is a different process with different costs, and it is outside what an email and a spreadsheet can solve.

A realistic week-by-week plan

• Week 1: log every wrong appearance, fix your own site, LinkedIn and databases, publish a clean company facts page. Cost: a few hours of staff time.
• Week 1 to 2: send correction requests for genuine errors, and update requests for stale facts. Cost: an hour or two per outlet.
• Weeks 2 to 6: submit sourced edits to Wikidata, Crunchbase and the knowledge panel, and pursue one or two fresh, accurate placements. Cost: low for edits, variable for placements.
• Weeks 4 to 8: re-run your original questions across search and assistants, compare to your screenshots, and chase only what is still wrong.

The position behind all of this is simple: treat a wrong fact as a source problem, not a chatbot problem. Fix the source you control, ask politely and precisely for the sources you do not, then outweigh what remains with better, newer evidence. It is less satisfying than a furious email and far more likely to work.