AI Search Tools Cited a Made Up Concept Within a Week

AI search citations can change within days of new content going live, which means whatever is published about your business or your subject can reach AI answers far faster than most marketing teams expect. That speed matters for anyone investing in AI search visibility for B2B businesses, because the same route that carries accurate information also carries errors, out of date details and claims made by people who have nothing to do with you.

The latest evidence comes from research reported by Search Engine Journal, which found that four English language AI search tools cited posts about an invented concept within a week of publication. Other test topics mostly failed to land. That contrast says a lot about how these tools choose sources.

What the Researchers Tested

The researchers published articles about a fictional idea on websites that AI search tools already cite heavily, then checked every day whether the tools repeated it. Their paper, study of public posts and AI search citations, analysed 17,211 citations across 10 AI search platforms before running the experiment. It found that 8 of the 10 platforms cited the fabricated concept within seven days.

The team first mapped which websites each tool cites most often, then tested how easy it was for a brand new account to publish on those websites. They wrote articles containing planted details, such as invented names and numbers, so they could tell whether an answer had drawn on their posts. Before publishing, they confirmed none of the tools gave an informative answer on the topics, so any later mention could be traced back to the test.

The team used three topics in the experiment. One was an invented concept, one was an opinion question about whether AI would replace junior programmers and one was a fictional keyboard. The tools named in the English language group were ChatGPT, Perplexity, Grok and Google’s AI search features. All four picked up the invented concept.

Why Some Topics Landed and Others Did Not

Topics landed when there was little else on the web to compete with them and when the posts sat on websites the tools already trusted. The invented concept had no competing sources at all, so the new posts were the only material available to answer questions about it. The opinion question and the fictional product had more competition and drew far fewer citations from the English language tools.

Placement mattered more than volume. The paper reports that a single article on a frequently cited website had more citation impact than over 20 matched posts on websites the tools rarely cite. The researchers also found that 40 posts on a self hosted website drew no references at all within seven days, according to the Search Engine Journal write up of the study.

The study has clear limits and the authors are open about them. None of the topics was a real brand or product, several results rest on small samples and the monitoring covered only one week per topic. The authors also say they could not separate the effect of the topic, the level of competition and the type of question, because all three changed together.

That last point is the one most relevant to an established business. A well known brand with years of consistent coverage is a very different case from an invented concept with no history. The businesses most exposed are those with thin coverage, a name that is easily confused with another organisation or details that have changed without the rest of the web catching up.

What the ChatGPT Citation Study Adds

A second paper looked at how ChatGPT chooses which pages to cite and found that matching was much stronger at the website level than at the exact page level. The study, Understanding ChatGPT Citations, found that Google results and OpenAI’s own API combined recovered only about one quarter of the URLs that ChatGPT cited, while doing noticeably better at matching the host website and the entity behind it.

The Search Engine Journal coverage of that paper gives the detail. For GPT-5.5 Instant, the combined sources matched 26.2% of cited pages but 48.7% of cited websites. The same write up notes that 44.5% of distinct pages for a question appeared in only one of 20 runs, which shows how much answers vary from one attempt to the next.

For a business, this means two things. AI tools tend to recognise trusted websites and organisations more consistently than individual pages, so the overall authority and clarity of your website counts for a lot. It also means a single check of what an AI tool says about you is not a reliable picture, because the answer and its sources can change between runs.

The Risk of Wrong Information About Your Business

The main risk is that an AI answer repeats something inaccurate about your business with the same confidence it gives to accurate facts. A separate guide on when ChatGPT has wrong information about your business covers the common causes. This research shows how quickly new material can feed those errors.

Most of these errors are not malicious. They tend to come from ordinary gaps that build up over time on B2B, professional services and public sector websites.

Those gaps matter more in light of this research, because the tools pick up whatever fills them so quickly. The common causes are listed below.

• Another organisation with a similar or identical name, whose details get blended with yours.
• Old addresses, phone numbers or leadership names still sitting in directories and press coverage.
• Services you have stopped offering that remain on archived pages or old PDFs on your own website.
• Third party profiles that describe your business in vague or inaccurate terms.
• Commentary on open publishing platforms that nobody at the business has seen.

The researchers also found that low cost posting packages could get content onto cited websites within minutes, which is a reminder that not everything an AI tool repeats was written in good faith. Google’s own guidance on AI features in Search says that seeking inauthentic mentions across the web is not as helpful as it might seem. Its spam systems also apply to the results its AI features draw on.

The defensive answer is not to flood the web with counter posts. It is to make sure the most authoritative and easily found source of facts about your business is the one you control. Every other public record should agree with it.

Make Your Website the Clearest Source About You

Your website should be the single most complete and consistent description of who you are, what you do and how to reach you. AI tools recognise trusted websites more reliably than individual pages, so an accurate, well organised website is the strongest signal you can send.

Start with an About page that states your registered company name, trading name, location, sectors and services in plain sentences. Keep service pages current and retire or redirect pages for services you no longer offer. Remove or update old PDFs, because tools can still find and cite documents that your navigation no longer links to.

Structured data helps machines connect your website to your organisation. Google’s documentation on Organization structured data explains how properties such as legal name, address, logo and links to your official profiles help it understand your administrative details. Google’s AI guidance is clear that structured data is not required for its generative features, so treat it as one part of a consistent picture rather than a fix on its own. Our guide to structured data for AI covers the markup in more detail.

Entity consistency matters as much as the markup. Your company name, address and description should match across your website, your Companies House record, your business listings and your social profiles. The public Companies House register is often one of the first places a tool or a researcher will check, so make sure your registered details and filings are current. Our article on entity SEO and AI visibility explains why this joined up picture helps.

How to Check and Correct What AI Tools Say About You

Checking what AI tools say about your business needs a repeatable process rather than a one off search. Because answers vary between runs, the aim is to spot patterns over time rather than react to a single reply.

Ask the main AI search tools the questions a buyer would ask, such as who you are, what you do, where you are based and how you compare with alternatives. Record the answers and the sources each tool cites, then repeat the same questions monthly. A standard AI SEO audit adds technical checks, such as whether AI crawlers can reach your key pages.

Fixing the source is where most of the effort goes. If an error traces back to your own website, correct it and request recrawling through your search console accounts. If it traces back to a third party, contact the publisher or directory with the correct details and evidence, such as your Companies House entry.

Name confusion needs its own approach. Where another organisation shares or closely resembles your name, state your registered name, company number and location clearly on your website. Use the same identifiers in your structured data. Clear differences such as sector, location and founding date give AI tools something concrete to separate the two.

Publishing original, useful content on your own website also helps over time. Google’s guidance favours first hand expertise over summaries of what others have said. A body of well written content marketing on your own domain gives AI tools accurate material to draw on when they answer questions about your sector.