Verify any claim · lenz.io
“Research by NP Digital found that more than 43% of marketers reported that false or hallucinated AI-generated information passed review and was published publicly.”
The conclusion
NP Digital published the quoted “over 43%” finding, but its study materials and related coverage do not present a consistent statistic. Most reports identify 36.5% as the share of marketers whose incorrect AI content reached the public, while 43.5% may describe an error category instead. The attribution is genuine, but the claimed percentage is not reliably established.
Caveats
- The primary source contains wording that conflicts with the figure most commonly reported from the same research.
- The 43.5% figure may describe a category of published errors rather than a percentage of all marketers.
- Repeated secondary reports may rely on the same NP Digital materials and are not necessarily independent confirmation.
Fact-check inside the tools you already use
Connect Lenz to ChatGPT, Claude, or WhatsApp and check a claim mid-conversation.
Or create a free account to bookmark this verification and run your own checks.
Sources
Sources used in the analysis
According to NP Digital’s AI Hallucinations and Accuracy Report, 47.1% of marketers encounter AI errors several times per week, while 36.5% say hallucinated or inaccurate AI-generated content has already gone live.
The report finds that 47.1% of marketers encounter AI errors several times per week, with 36.5% saying hallucinated or inaccurate AI-generated content has gone live. … More than one-third of marketers (36.5%) report that hallucinated or incorrect AI content has been published publicly, most often due to false facts, broken citations or brand-unsafe language.
In our research, over 43% of marketers say hallucinated or false information has slipped past review and gone public.
Over 70% spend one to five hours fact-checking AI each week, and more than a third (36.5%) say incorrect AI content has gone live publicly.
Nearly half of marketers (47.1%) encounter AI inaccuracies several times per week, and 36.5% report that hallucinated or incorrect AI content has already gone public.
In our research, over 43% of marketers say hallucinated or false information has slipped past review and gone public.
More than one-third of marketers (36.5%) admitted that hallucinated or incorrect AI-generated content has already been published publicly, according to the findings released February 2, 2026. … The most common offenders included inappropriate or brand-unsafe content (53.9%), false or hallucinated information (43.5%), and formatting glitches that broke user experience (42.5%).
The report finds that 47.1% of marketers encounter AI errors several times per week, with 36.5% saying hallucinated or inaccurate AI-generated content has gone live. … - AI Errors Made Public: More than one-third of marketers (36.5%) report that hallucinated or incorrect AI content has been published publicly, most often due to false facts, broken citations or brand-unsafe language.
According to NP Digital’s newly released AI Hallucinations and Accuracy Report, 47.1% of marketers encounter AI errors several times per week, and 36.5% say inaccurate or hallucinated AI-generated content has gone live publicly.
The tasks that fail most are the ones marketers rely on AI for most heavily: HTML and schema creation (46% daily error rate), full content development (43%), and reporting (34%). … Despite this, 36.5% admit AI errors have reached the public, and another 39.8% say they've caught close calls right before publishing.
Continue your research
Verify a related claim next.
Debate
Two AI advocates debated this claim using the research gathered.
Argument for
NP Digital's own data-backed study explicitly states that “over 43% of marketers” said hallucinated or false information slipped past review and went public, precisely matching the atomic claim (Source 3, NeilPatel.com). This is a direct, verbatim, verified statement of the research finding, so it establishes that NP Digital reported the claimed result.
The Proponent cherry-picks a single, poorly phrased blog post (Source 3) while ignoring the overwhelming consensus of official press releases and independent reporting which confirm that only 36.5% of marketers published AI errors (Sources 1, 2, 8). By relying on this isolated misstatement, the Proponent fails to recognize that the 43% figure actually represents the frequency of false information among those published errors, not the percentage of marketers who published them (Source 7).
Argument against
The claim misrepresents the data, as a broad consensus of independent reports confirms that only 36.5% of marketers admitted that hallucinated or incorrect AI-generated content was published publicly (Sources 1, 2, 4, 5). Furthermore, the 43% figure actually refers to the proportion of published errors that consisted of false information, rather than the percentage of marketers who published them, demonstrating a clear misreading of the statistics (Source 7).
The Opponent commits a category error by substituting the report's separate 36.5% measure of any hallucinated or incorrect content going live for the distinct finding that over 43% of marketers said false or hallucinated information passed review and went public, which NP Digital states verbatim in its own study (Source 3, NeilPatel.com). The Opponent also misreads Source 7: its 43.5% figure concerns false or hallucinated information among error types, but that does not contradict, qualify, or disprove the directly reported marketer-response figure in Source 3.
Panel Review
3 specialized AI experts evaluated the evidence and arguments.
Reviewer 1 — The Logic Examiner
Source 3 directly and verbatim attributes the claimed over-43% finding to NP Digital, but Sources 1, 2, 4, 5, and 8 consistently report 36.5% for marketers whose hallucinated or incorrect content went public; without evidence that these are genuinely different survey questions or denominators, the higher figure is in unresolved tension with the broader reporting, while Source 7 does not by itself establish the opponent's alternative interpretation. The claim is therefore mixed: it accurately repeats NP Digital's own published wording, but the available evidence does not logically reconcile that wording with the repeatedly reported 36.5% result sufficiently to establish the asserted finding as reliable.
Reviewer 2 — The Source Auditor
The primary source itself (Source 3, neilpatel.com, the actual NP Digital study) explicitly and verbatim states 'over 43% of marketers say hallucinated or false information has slipped past review and gone public,' matching the atomic claim precisely and duplicated in Source 6 (staging mirror). However, the majority of independent secondary reporting (Sources 1, 2, 4, 5, 8, 9 - Adweek, GlobeNewswire, Media in Canada, MarTech Edge, Ethical Marketing News, Marketing News Canada) consistently cite 36.5% as the figure for content going live publicly, and Source 7 (ppc.land) clarifies that 43.5% refers to false/hallucinated information as a category of error type among published content, not the percentage of marketers reporting publication—suggesting the neilpatel.com blog post's '43%' framing may be an internal inconsistency or loose paraphrase within NP Digital's own materials rather than the headline finding replicated everywhere else.
Reviewer 3 — The Precision Analyst
The claim states that over 43% of marketers reported false or hallucinated AI-generated information passed review and was published publicly. However, multiple sources (1, 2, 4, 5, 7, 8, 9, 10) clarify that 36.5% of marketers reported incorrect AI content going live, while the 43.5% figure refers to the proportion of those errors that were false or hallucinated information (Source 7), meaning the claim conflates the two statistics.
Panel summary
NP Digital's own article explicitly states that “over 43% of marketers” said false or hallucinated information passed review and went public, directly supporting the attribution. However, most coverage of the same research reports 36.5% as the share of marketers whose incorrect AI content went live. Some reporting associates 43.5% with the type of error involved, although the available wording does not conclusively establish that interpretation. Because the underlying materials contain an unresolved and material quantitative inconsistency, the precise 43% finding cannot be treated as reliable despite appearing verbatim in the primary source.