Recommended Under a Name You No longer Have

A buyer who follows an AI recommendation expects the firm at the other end to exist. In this study, about one answer in three recommended a firm that exists, does the work, and has not operated under the recommended name for years.

The three firms involved are different sizes and changed their names for different reasons. The way the retired names survived is different in each case too, and the differences are the useful part.

The firm whose own website keeps the old name alive

One of the ten firms on our panel renamed itself in April 2024. The old name's registration with the state expired that May. Seventeen months later, on both of our collection dates, the AI assistants recommended the firm under the old name 62 times, then 61, and under its current name zero times. We read all 61 of the September mentions by hand. Every one of them presents the old name as a firm to hire today: a numbered recommendation, a bullet with a "why consider them" line, no "formerly," no "now known as." The current name does not appear anywhere in any of the 61 answers.

In August, every one of those recommendations came from the three assistants that answer from memory. That is the expected pattern. The models learned the market as it stood when they were trained, and the firm had the old name then. There is nothing a firm can do about that quickly.

In September something changed. Google AI Mode, which searches the live web and cites its sources, recommended the firm under the old name too: one answer to our Durham child-and-family-health question, returned five times. We read it in full. The firm is the first of four bullets, with a town and an endorsement. The sources Google cited were the firm's own current website, at the domain it adopted with the new name, plus its LinkedIn company page. So we went and looked at those pages.

The homepage title tag, the text a search engine and an AI reader see first, read "[Old name] - Trusted Advisors for Over 50 Years." The audit-and-assurance page Google cited is titled with the old name. The team page describes a principal as part of "the [old name] team since 2005." The LinkedIn company page has never been renamed. The firm changed its domain. It did not change the words on it.

That is the finding in one sentence: a retrieval-based assistant read the firm's own site and recommended it by the name its own site still uses. The federal record, incidentally, shows the rename clearly; the firm's filings carry the old domain through 2023 and the new one from 2024. The machines were not reading the federal record.

A buyer who takes that recommendation will search for a firm that, on paper, ceased to exist in May 2024. Some will find their way to the right door. Some will not.

The firm whose old name survives only in memory and in the state's files

A second panel firm, headquartered outside North Carolina, acquired a Durham firm by merger in June 2023. The Durham firm had operated under its own name since 2001. We did not go looking for this one; we found it by text search after correcting an unrelated error in the firm's website address, and the fifteen mentions were not in our scoring because the instrument's alias list did not include the predecessor name. We report them as found, and we read all fifteen by hand.

Fifteen of fifteen present the Durham firm's old name as a current firm to engage: "Located right in Durham," "Best for," numbered recommendations with specialties attached. None of the fifteen mentions the acquiring firm. The acquiring firm itself, under its own name, was recommended once in August and never in September.

Thirteen of the fifteen come from the memory assistants. The two that cite sources, both from Perplexity in August, cite the North Carolina State Treasurer's audit-fee data, which still lists the Durham firm as auditor of a Durham charter school. So the old name survives in two places: inside the models, from training, and inside a state agency's public records, from the years the firm did the work under that name. The predecessor's own website is gone; its domain now shows a site-builder placeholder. There is nothing of the firm's own left for a machine to read, and nothing of the successor's that a machine chose to read instead.

In one ChatGPT answer in September, this firm's retired name and the first firm's retired name sit in the same numbered list of recommended firms.

The national firm recommended under both halves of a merger

A national firm formed in 2022 from the merger of two large regional firms rounds out the count. It was mentioned under one predecessor's name 97 times in August and 89 in September; in roughly a third of those, read by hand, the old name is presented as a current firm with no mention of the merger. All of those came from one assistant. In one answer, the two predecessor firms are listed one after the other as separate firms to consider: both halves of the merged firm, under the names they gave up four years ago.

Three mechanisms, one count

Added together: 79 of 250 answers in August and 85 of 250 in September presented a firm under a retired name as a firm to hire today. Both figures were hand-read.

The three cases run on different mechanisms, and that matters for what a firm can do:

• The rename. The old name survives on the firm's own current pages and its LinkedIn profile. A retrieving assistant read them and repeated it. The memory assistants also hold the old name, but the retrieval problem is the firm's to fix, this month.
• The merger, small. The old name survives in the models and in a state agency's records. The predecessor's site is gone. The successor's site exists and was never read for these questions, for reasons covered elsewhere in the study (its service pages name no place). The firm cannot edit the Treasurer's files or the models; it can only publish something of its own that is newer and reachable.
• The merger, national. The old names survive in the models alone, four years on. Nothing on the web is feeding them; nothing on the web has displaced them either.

What a visibility score would say about all three: the firm is "mentioned." What a buyer experiences: a recommendation for a firm that does not pick up the phone under that name.

What a renamed or merged firm should check

1. Search your own website for the old name. Not the homepage only. Title tags, page titles, team biographies, image alt text, PDF footers. In this study the homepage title was the first thing a machine saw and it was the old name.
2. Check your LinkedIn company page. It was cited as a source. Company pages are renamed by request to LinkedIn, not by editing, and many firms never do it.
3. Ask an assistant about your old name. If it describes a current firm with your specialties, that is what a buyer who heard the old name from a colleague will be told.
4. Look for your old name in public records that machines read. State treasurer audit-fee data, school board minutes, grant databases. You cannot edit them. You can make sure that a search for the old name finds a current page of yours that says what happened.
5. If you acquired a local firm, give its old name a home on your site. One page: "[Old name] is now part of [your firm]," with the office, the people and the work. The machines will keep saying the old name for as long as the models do; a page that catches the buyer when they look it up is the one thing within reach.

What this does not show

Not why any assistant chose a name; correspondences between sources and names are observations on two dates. Not the state of either firm's site on 11 August; the technical read was taken after the second collection. Not a rate: 79 and 85 of 250, on two dates. The fifteen predecessor-name mentions of the merged panel firm were found after the fact and were not part of the pre-registered scoring; they are reported as a post-hoc finding and labeled as such on the record page.