Dissolved but still recommended: AI engines named 7 of 16 closed UK companies again

Last Updated on September 17, 2026

Last checked: 15 September 2026. Register re-check and repeat questions both run on 15 September 2026. Counts only. This page names no businesses.

Whito has already shown that AI engines sometimes recommend businesses whose company has closed. This study asks the next question. Once an engine has named a closed company, does it keep naming it?

We took 197 business names that three earlier Whito collections had logged. We checked every one against Companies House again on 15 September 2026. Then we put the same question to the same engine for the same town and trade, and recorded which names came back.

Sixteen names matched only a company that was dissolved or in liquidation. Seven of them were named again. Five of those seven were the first business in the answer or were named in its opening sentence.

What we did

The names. Three sets, all collected and logged by Whito before this study:

The second set was chosen because those names were already known to be closed or close to it in August. It makes the closed group big enough to count. It also means this study does not estimate how common closed companies are in AI answers. For that, see our 728-name verification study.

The 20 towns are Reading, Sunderland, Swansea, Leeds, Nottingham, Coventry, Luton, Bristol, Doncaster, Ipswich, Preston, Stoke-on-Trent, Milton Keynes, Norwich, Plymouth, Glasgow, Belfast, Mansfield, Blackpool and Wolverhampton, so the AI Mode sets cover England, Wales, Scotland and Northern Ireland. The Manchester set is England only. Companies House is a UK-wide register. In Manchester, 104 mentions came down to 100 names because both engines gave four of them.

Step 1: register re-check, 15 September 2026. Every name was searched on the Companies House public register. For every plausible match we opened the company record and read its status, registered office and filing flags. For every active match we read the type of its latest accounts, to catch companies filing dormant accounts. We then ran 39 further searches, with other spellings and former company names, on 28 of the names. That is four steps in all.

Each name got one result:

Our rules, in order. A company whose name matches exactly (ignoring case, punctuation and “Ltd”) decides the result if it is in or near the town. Where two companies share that exact name, an active one wins, and we flag the pair. Where there is no exact match, the nearest active company carrying the same distinctive name decides it, flagged as a close call. A name made only of a town and a trade, such as “[Town] Roofers”, is too generic to match unless a company carries it exactly. A branch of a national brand counts as active when the brand’s own companies are active.

Step 2: the same question again, 15 September 2026. We asked 52 questions, word for word as before: 36 on Google AI Mode (the 20 roofer towns plus the 16 questions behind the second set), 8 on ChatGPT and 8 on Gemini. For each name we recorded whether the same engine named it again. A business that appeared only as a source label, not as a recommendation, did not count.

What we found

Table 1. Register result on 15 September 2026, by set (names)

Sixteen names matched only a closed company: 14 dissolved and 2 in liquidation. None matched a company filing dormant accounts.

Table 2. Were they named again? (15 September 2026, same engine, same question)

Table 2 counts each engine’s mention once, so a Manchester name given by both engines appears twice. All 16 closed names came from a single engine each: 15 from Google AI Mode and 1 from ChatGPT.

Table 3. Google AI Mode only

Closed names came back less often than other names. They still came back. On this evidence, a company closing does not take its name out of an AI Mode answer.

What the seven repeat answers looked like.

• Five of the seven were the first business named, or were named in the answer’s opening sentence.
• Five of the seven had been dissolved for more than a year. The oldest dissolution dates from February 2010.
• One answer gave, as the firm’s address, the registered office of a company dissolved in 2017. The same answer told readers to beware of namesakes.
• One answer called a cleaning firm “13+ years established”. The company that fits that name was dissolved in May 2026.

Other things the re-check found (counted once per name):

• 6 matched companies showed an overdue accounts or confirmation statement flag.
• 1 active company showed an active proposal to strike off.
• 2 active companies share their exact name and address with a company of the same name that was dissolved earlier.
• 4 names given by the engines are former names of companies that have since been renamed.
• No company moved between active, dissolved and liquidation since our earlier checks (28 August for the AI Mode sets, 14 September for Manchester).

Corrections to our own earlier checks. One roofer our 28 August check recorded as filing dormant accounts has filed micro-entity accounts in both of its last two years. It is now recorded as active. In Manchester, this pass settled all 13 names our 14 September check left open. Two other Manchester names, one practice given by both engines, are now the ones we could not settle. The September roofer page’s one-pass check found no company match for 36 of the 79 roofers. This fuller pass finds no active match for 31, plus 2 with a different registered name.

Six of the 18 names in the second set are now recorded differently from our 28 August check, although their company records have not changed. Two are names made only of a town and a trade, and two closed companies are registered too far from the town, so we no longer tie those four names to them. One is a national brand still trading through its group companies. One matches an active company that used to carry its name.

By engine, across all names: Google AI Mode named 61 of 97 again, ChatGPT 48 of 75 and Gemini 12 of 29. ChatGPT returned the same 12 Manchester roofers as the day before.

What this does not prove

• The closed group was picked on purpose. Twelve of the 16 closed names came from a set chosen because it was already closed in August. This study measures whether closed names come back, not how many AI answers contain them.
• One repeat question is one snapshot. Answers change from day to day and from account to account. The AI Mode repeat came 24 days after the first answer. The Manchester repeat came one day after.
• Accounts can shape answers. AI Mode and Gemini were used signed in. ChatGPT was used on the free tier, signed in to the same account as on 14 September, with those earlier chats still in its history.
• Two AI Mode answers were generated twice. For two roofer questions, one of which holds a closed name that came back, our first capture read the page before the answer appeared, so the question was asked again the same day. The second answer is the one recorded. Some Gemini questions needed a second try before they were submitted. Only one answer was produced for each.
• Matching is a judgement. The method changes the answer. On the same 728 August names, a single name search left 58% unmatched and a four-step check left 5.2% unconfirmed (how accurate are AI business recommendations). Close calls are flagged in our file. Two of the six “different registered name” results rest on a link made in our 28 August check.
• “No active match found” does not mean fake. A sole trader registers with HMRC for Self Assessment and does not appear on Companies House as a company. Many honest firms trade under a name that is not their company name.
• Dissolved does not always mean the business has stopped. The same people may trade as a sole trader or under another company. A company struck off and dissolved within the last six years can apply for administrative restoration. What the register does show is that the company the name points to is closed.
• A blank field proves nothing. An empty field on Companies House or any other register does not show that something is missing in real life.
• This is not a ranking. We report which names were given, by which engine, on which day, and what the register showed. We do not say which firms are good or bad.
• The sample is small. 52 questions, three engines, one day. It does not represent every UK town or trade.

What a business should do next

If you run a business, check your own record. It takes five minutes.

1. Search your trading name on the Companies House register. Make sure the company that comes up first is yours and is active.
2. If an old company of yours is dissolved but still carries your trading name, make it easy to find the live one. Put your current registered company name and number on your website.
3. If you renamed your company, update the name on your website, your Google Business Profile and the directories you use, so the old name stops travelling.
4. Check what search engines and AI assistants can read on your homepage with the free Whito Visibility Check. It runs 17 automated checks. No email needed.

Free Visibility Check: https://whito.co.uk/visibility-check/

If you are hiring, search the name on Companies House before you pay a deposit. Being named by an AI engine, even first, does not show that the company behind the name is still trading.

This page is not legal advice. If your own company has been dissolved and you are still trading, speak to an accountant or solicitor.

The file behind this study

The row-level file holds all 201 mentions: engine, town, trade, company number, register status as read on 15 September 2026, match confidence, flags, notes, and whether each name came back. It is available through Whito Data. Nobody can pay to be added to this study, removed from it or moved within it. How Whito makes money.

Cite this research

Whito Research (2026). Dissolved but still recommended: AI engines named 7 of 16 closed UK companies again. https://whito.co.uk/research/dissolved-companies-still-recommended-by-ai/

If your business is in our data and a record is wrong, tell us through corrections. We check it against the register and our logs, and we record any change.

Related Whito research

• AI-recommended roofers, checked against Companies House (79 names, September 2026)
• When two AI engines pick a trustworthy local firm, they rarely pick the same one (Manchester, 14 September 2026)
• How many UK businesses AI calls trustworthy are not a real, active firm in that town (728 names, 28 August 2026)
• We asked AI which electrician to hire (182 names, July 2026)
• The Whito AI Visibility Index, Q3 2026

Jacob Whitmore Founder and Lead Analyst

Jacob Whitmore founded Whito. He works on what UK businesses are actually charged for marketing, and on getting those numbers published with the date they were checked.

See Full Bio