Our AI search tool invented a new 2026 Depeche Mode album that does not exist
A technical look at how an AI web-search tool fabricated a nonexistent Depeche Mode album during Side-Line’s own AI-assisted research on September 16, 2026.
During Side-Line’s own AI-assisted research today, September 16, 2026, an AI web-search tool returned a fabricated claim: that Depeche Mode was releasing a new studio album titled “Come This Madness” on September 25, 2026. No such album exists, of course. It was another AI agent — one we developed to check the claims made by other AI tools — that prevented the article from landing in our editorial flow for further work.
Since many of you reading this work with AI professionally, we just wanted to give you a technical account of how that specific kind of error happens. It also shows why all publications need to invest in verification when working with AI. And as an extra, here’s a very suitable Depeche Mode video to go along with this article :).
The tool searches, then writes a very short summary for us to work on
The AI search tool in question does not work like a traditional search engine that returns a ranked list of links. It runs a query, retrieves a set of web pages or snippets that match it, and then uses a large language model to generate a written summary of what those results say. This two-step process, retrieval followed by generation, is standard for AI-assisted search and is usually described as retrieval-augmented generation.
The retrieval step worked correctly in this case. The underlying search returned real, relevant results, including a legitimate Mojo magazine feature on Depeche Mode’s Berlin period and coverage of the band’s 40th-anniversary “Black Celebration” material. The failure happened in the second step, when the language model converted those results into prose.
Why a language model can state a false fact with confidence
Here it gets interesting, of course. A large language model generates text one token at a time by predicting the statistically likely continuation of the text so far, based on patterns learned from its training data. When it is asked to summarize search results, in this case on Depeche Mode, it is not running a database query that returns exact matches; it is composing a new sentence that resembles the kind of sentence that would plausibly follow from the material it has been shown, blended with patterns from unrelated text it saw during training.
For a band with a five-decade recording history, dozens of studio albums, tours, and reissues, and a large body of fan and press writing, the model has an enormous internal store of patterns for what a Depeche Mode album announcement looks like: a title, a release date roughly a few weeks out, a label. When the retrieved search results did not contain a clean, direct answer to “Depeche Mode news September 2026,” the model filled the gap with a statistically plausible-sounding invention rather than stating that no such album was found.
This behavior is commonly called a hallucination or, more precisely, a confabulation: fluent, well-formed text that is not grounded in any specific retrieved source, generated because the model’s training optimizes for producing plausible continuations, not for flagging uncertainty.
No per-sentence source attribution
A second technical factor made the error harder to catch at the point of generation. The summary presented several genuine facts, drawn from real search results, alongside the fabricated album claim, in the same paragraph and the same tone, with no inline citation marking which sentence came from which source. When a retrieval-augmented system does not attach a verifiable source link to every individual claim in its output, a fabricated sentence is indistinguishable, on the page, from a verified one. Some AI search tools address this by forcing every sentence to carry a citation and refusing to generate sentences that cannot be tied to a specific retrieved passage; the tool used here did not do that consistently for this query.
How the Depeche Mode error was caught
Our editorial workflow treats AI-generated claims as unverified until checked against a primary or otherwise reliable source. The AI agent we use for this directly checked official Depeche Mode sources, and none mentioned any album by that title. A separate search of Bing (as a double check for Google) for the same underlying query surfaced no matching result anywhere on the public web, confirming that the claim had no traceable origin.
Because it could not be sourced, it was discarded and the Depeche Mode article pitch never ended up in our flow. We found it while routinely checking the results of the AI agents at work to see if the agents were doing their work properly. You have to do this as an AI operator, or you end up with some pretty weird results, as this example shows.
Also, you might want to know that we run various AI agents (each specialized in certain data) to catch factual errors in promo copy that we get (you would be surprised how many press texts sent by labels and bands have factual errors) before we start working on the articles themselves.
Support Side-Line
More people read Side-Line than ever, and there is still no paywall and no advertising. Readers pay for it instead - from $5 a month.
Why we ask
Advertising revenues across the media are falling fast. Unlike many news organisations we have not put up a paywall, and we refuse to add annoying advertising, so we need to ask for your help. Side-Line’s independent journalism takes time, money and hard work to produce, and we do it to push the artists we like, who are equally fighting to survive.
Handled by PayPal.
Chief editor of Side-Line – which basically means I spend my days wading through a relentless flood of press releases from labels, artists, DJs, and zealous correspondents. My job? Strip out the promo nonsense, verify what’s actually real, and decide which stories make the cut and which get tossed into the digital void. Outside the news filter bubble, I’m all in for quality sushi and helping raise funds for Ukraine’s ongoing fight against the modern-day axis of evil. Besides music I’m also an SEO and AI content flow specialist and have an interest in everything finance from stocks to crypto. There is music in everything!
