Specifically, I asked about Conduct™ — one of our published software concepts. Here's what it is, so the gap between that and what Gemini said lands properly.
Conduct™ is a design concept for iOS, iPadOS, and macOS that reimagines how notifications work at a systems level. Its central argument: Apple's Focus treats notifications like a volume knob. Conduct treats them like people. The system is built around the idea that a human day has a shape — Morning, Work, Evening, Sleep — and that human relationships have a structure. Rather than toggling apps, Conduct routes access by relationship and time period simultaneously. One decision per person covers calls, messages, and email at once. A notification architecture built around how people actually live.
iOS ecosystem. Human attention, human relationships, the shape of a day. Publicly documented. No connection to enterprise software, data pipelines, CRMs, or invoice platforms. Not a data integration tool. Not middleware.
Here's what Gemini said it was:
"designed to solve 'The Coherency Problem' — the frustration of moving data between isolated tools like a CRM and an invoice system that don't speak the same language, which he calls the 'Copy-Paste Tax.'"
None of that exists. Not the problem name. Not the terminology. Not the product category. A completely different company, under our name, described with total fluency and zero accuracy.
I tried the premium thinking model — the one that charges more because it reasons before responding. It invented a more elaborate version. Added departments, connective tissue, enterprise workflows. More reasoning, more confidently wrong.
What makes this stranger than a typical AI error is that Gemini didn't always get this wrong.
Two months ago, the same query returned something genuinely impressive. Accurate business model. Correct workshop methodology. Right consulting approach. The kind of summary that would take someone hours to produce from scratch. It had real, meaningful context about what we do.
So this isn't about being a small company that AI doesn't know about. It had accurate information. Something changed between January and March — silently, with no warning, no changelog, no degradation notice. The ground just moved.
Here it is in full context, not just the excerpt above.
"Klaiber has recently been developing Conduct, a software concept designed to solve 'The Coherency Problem.' It addresses the frustration of moving data between isolated tools (like a CRM and an invoice system) that don't speak the same language, which he calls the 'Copy-Paste Tax.'"
Every single factual claim in this paragraph is fabricated. "The Coherency Problem" is not a Conduct concept. "Copy-Paste Tax" is not our terminology for Conduct. CRM systems and invoice platforms have zero connection to what Conduct does. Gemini described — in confident, specific terms — a product that does not exist, and attached our name and brand to it.
I switched to the premium model. The one that reasons before it responds. Asked about our other concept apps.
"Klaiber has been developing a software concept called Conduct. It is designed to solve what he calls 'The Coherency Problem' — the fact that as companies grow, their tools stop agreeing with each other. The Goal: To create a platform that acts as a 'connective tissue,' ensuring that data remains consistent as it moves through different departments, eliminating the need for manual workarounds."
More thinking produced more elaborately wrong output. The fabricated terminology expanded. The hallucinated use case deepened into an invented enterprise platform with "departments," "connective tissue," and "manual workarounds." The extended reasoning model didn't correct the fast model — it amplified the fiction, adding structure and detail to something wholly invented. Then it failed to mention a single one of our four other concept apps that actually exist and are publicly documented.
When asked about our other concept apps, the thinking model recycled its hallucinated Conduct description and listed nothing else. Here's what it completely missed — five published, publicly accessible concepts, all documented on our workshop page:
Relationship-first notification management for iOS/iPadOS/macOS. Not a CRM. Not middleware.
Pivot-filtered email for inbox clarity. Single-click sender or subject filtering without leaving the inbox.
Microsoft Teams reimagined around precision messaging. Time, intent, and accountability stay visible.
Context anchoring, device hierarchy, per-app budgets. Three fixes that should have shipped a decade ago.
Persistent reference windows for macOS. One toggle that protects focus and eliminates window-switching tax.
Every single one is about human attention, workflow, and the UX of operating systems and productivity tools. Not one is an enterprise data integration platform. The portfolio is findable. The concepts are documented. Gemini invented a sixth company instead.
Want more proof? See how it completely butchered MMWBMail too →
MMWBMail is a pivot-filtered email inbox concept — single-click sender or subject filtering without ever leaving the inbox, built around user intent rather than platform engagement. Here's the biography Gemini invented for it instead: a fake domain, invented client services, and a fabricated technical architecture across seven bullet points.
Called it a "specialized internal tool and integration layer" — not a user-facing UX concept. An enterprise back-end integration tool.
Named it a "data-aware communication hub" built around a framework called "Coherent Communication." Neither phrase exists in MMWBMail.
"Contextual Sourcing" — pulling live data from CRMs, billing, and project management tools into the email composition window. Prevents the "Shadow System" problem. Fabricated feature. Fabricated problem name.
"The One-Way Trip for Data" — a philosophy that data entered in email flows back automatically to a central source of truth. Does not appear anywhere in MMWBMail.
"Smart Fragments" — reusable data blocks always up-to-date and formatted for the recipient's device. A fabricated feature name for a fabricated feature.
Claimed it is used as part of a "10-day diagnostic" client service bridging "the fracture" between email and core business operations. No such service is tied to MMWBMail.
Stated the domain mmwbmail.de hosts "high-security, private integrations" where Klaiber manages "connective tissue" between a client's siloed tools. The domain, the integrations, the service model — all fabricated.
Two products. Two completely different fabricated identities. The same fluency. The same specificity. The same confidence. This is a pattern, not an accident.
So there's no ambiguity about how complete this failure is:
Conduct doesn't address a "Coherency Problem." Its argument is that Apple's notification model lacks a coherent theory of people and time. Completely different domain. Not our terminology.
Conduct has zero to do with data movement, CRMs, or invoicing. It's a personal notification concept for Apple platforms. Wrong category by several hundred miles.
Not a Conduct concept. Not our terminology for this product. Sourced from somewhere unrelated or manufactured outright, then confidently attached to a product it has nothing to do with.
Conduct is a personal iOS notification system. There are no departments. No organisational data. The thinking model invented an enterprise B2B platform and put our name on it.
Gemini did not mischaracterize Conduct. It did not get a feature wrong or confuse a detail. It described a completely different product, in a completely different market, solving a completely different problem, using completely different terminology — with the fluency and specificity of something it had actually researched.
A consultancy that helps other companies get their software right, being misrepresented by the biggest AI on the planet. That's not just embarrassing for Google, it's a little funny.
Less funny is what this means if you're not in a position to know the answer is wrong.
The danger isn't that AI occasionally gets things wrong. It's that it gets things wrong in ways that are indistinguishable — on the surface — from things it gets right.
This is where the story stops being about Gemini and starts being about how businesses are operating right now.
Every day, business owners use AI to research competitors, brief teams, write proposals, validate ideas, and make decisions — without any systematic process to verify what comes back. The failure mode isn't the AI saying "I don't know." That would be manageable. The failure mode is the AI saying something completely wrong in the confident register of something completely right.
A prospect asks an AI about your company before a call. The AI describes someone else's product under your name. The prospect arrives confused — or doesn't arrive at all. You never know why.
Teams use AI to brief themselves on market context, competitive positioning, and partner capabilities. If that information is fabricated, every downstream decision inherits the error.
The extended reasoning version didn't catch the error — it deepened it. Businesses paying for premium AI tiers may be getting more elaborately confident hallucinations, not safer outputs.
Two months ago this worked. Today it doesn't. No notification. No changelog. No degradation alert. Models shift, training data changes, and the ground can move beneath you without a sound.
Stop treating AI output as a research artifact. Treat it as a first draft that requires verification — every time, for anything that matters. Build surfacing into your workflow. Test regularly. Verify against primary sources.
What Gemini said about Conduct isn't a scandal. It's a case study in what happens when the verification layer is missing.
If you're putting your business, your brand, or your client relationships on the line based on AI output you haven't checked — this is exactly what you're betting on.