Product1 distinct publisher3 min readUpdated
Which chatbot drew the map is the least useful question. The reproducible defect is a bound print run with no human proof step, on an artifact that cannot be patched after delivery.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
Which chatbot drew the map is the least useful question. The reproducible defect is a bound print run with no human proof step, on an artifact that cannot be patched after delivery.
Farnsley Middle School in Kentucky sent students home with a bound agenda containing AI-generated graphics that were never corrected: the map of the United States labels Illinois "Vitoiis" and Kentucky "Venecky", the solar system picks up an extra planet called Marc and renames Saturn "Sature", and the periodic table is, in Gizmodo's description, nonsense [1][2][4][3]. The story was first reported by the Louisville outlet WDRB [5]. What matters here is not which generator produced the artwork but that a printed, bound object reached students without anyone reading a page of it.
Gizmodo's own question was which chatbot was used, and it ran the test [6]. ChatGPT, Grok and Claude all returned accurate maps, with Claude using state abbreviations rather than full names, and all three produced periodic tables that looked correct [7]. Google's Gemini, which uses Nano Banana for image generation, produced errors of the same family, including a "Nontana" west of Montana and a "Sork" coast [8]. So three of the four tools tested cleared the bar [9]. Gizmodo says it still does not know what was used at Farnsley, and floats older image generators or an obscure tool sold to the district as a specialised package [10].
That uncertainty is the point. The generator is a variable that changes every few months; the absent proof step is the constant. A bound agenda has a hard commit: once it is printed and handed out, there is no hotfix, no silent revision, no rollback. Any workflow that puts unreviewed model output into a print queue is a workflow that will eventually ship something like this, and the choice of vendor only changes how embarrassing the specific artifact is.
There is also evidence the gap is not one person's slip. The district's website carries an AI statement saying it views GenAI "not as a replacement for human intellect or the invaluable role of educators, but as a powerful tool to enhance learning, teaching, and operational efficiencies" [11]. The same website, according to Gizmodo, is illustrated with AI images showing children with too many fingers, phones with nonsensical designs, and tablets hovering in front of students [12]. Published policy and published output disagree, which usually means no one owns the review gate.
The district has not closed the two questions an operator would ask first. WDRB got no proper explanation beyond a statement that the district is communicating with parents, and no answer on whether the problem was confined to one school or ran across the district [13]. Gizmodo says the district did not respond to questions emailed Tuesday afternoon [14]. Scope determines the fix: one teacher's file is a training problem, a shared template or a purchased package is a procurement and sign-off problem. Nor is education alone in this. Gizmodo notes the U.S. State Department recently presented a map of Africa with country names placed incorrectly [15].
Watch for a scope answer and a named tool from the district, and for any change to who signs off before a print run. Watch, too, whether the illustrations on the district's own site are replaced, since that is the cheapest available signal that someone now checks the output [12].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
OpenAI's ChatGPT, xAI's Grok and Anthropic's Claude all generated accurate maps of the U.S. with names, though Claude used abbreviations rather than full state names; the periodic tables created by each also looked accurate.
A school in Kentucky sent students home with a bound agenda that included absurd errors clearly made by AI.
The agenda's map of the U.S. lists the state of Illinois as "Vitoiis" and Kentucky as "Venecky".
The agenda includes an extra planet called Marc, and Saturn is listed as "Sature", among other gibberish errors.
The story was first reported by local Louisville news outlet WDRB, concerning the Farnsley Middle School agenda.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One secondary account of one local report
The artifact errors are concrete and specific, and Gizmodo attributes the original reporting to WDRB, which had video of the pages. But the cluster contains a single source, the district has not confirmed anything, no vendor or workflow record exists, and the tool attribution is openly unknown. The model comparison that anchors the headline is an undocumented single run.
One district, unknown scope
There is documented generative-AI use inside one Louisville-area district: a printed agenda distributed to students plus AI imagery on the district website. Nothing establishes how many agendas, how many schools, or any wider institutional pattern, and the district refused to say whether the defect was isolated. Adoption is therefore real but narrow and unquantified.
Framing outruns attribution
The documented defect — unreviewed generated content in an unpatchable print run — is understated nowhere and holds up. The overstatement is in the framing: the headline poses a chatbot whodunit that the article cannot answer, and a single unreplicated test is used to single out Gemini/Nano Banana while three rivals pass. The State Department Africa map is cited as pattern evidence without connection to this incident.
Virality pull on one side, silence on the other
Gizmodo has a clear engagement incentive in an AI-slop story: mocking asides, an invitation for readers to name the culprit chatbot in comments, and a model-comparison hook that generates a named villain. On the other side, the district's incentive is reputational containment, which shows up as no substantive answer to WDRB and no reply to Gizmodo's emails. Both pressures push the account toward spectacle and away from process facts.
Core artifact solid, causation soft
High confidence that garbled AI-generated graphics reached students in printed form and that the district has not explained cause or scope. Low confidence in any claim about which tool was used, in the relative reliability of specific image models, and in how widespread the practice is — all resting on one publisher, one informal test, and no institutional confirmation.
product
Incogni ranks 13 AI assistants by privacy risk: bigger is worse, except ChatGPT1 distinct publisher
security
The nationalization argument is really a vendor-continuity memo1 distinct publisher
build
Count invalid JSON as a failed classification, and model choice becomes a reliability problem1 distinct publisher
build
The Tokenizer Is Your Real Price List, Not the Per-Million Rate Card1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 18, 2026