Some AI products let you talk to a simulated version of a historical scholar. Type a question, get an answer in the voice of Al-Ghazali or Ibn Taymiyyah. It is a compelling idea. It is also one that can go wrong quietly, in a way most users would never notice.

Here is the problem in plain terms. A language model does not know what Al-Ghazali would have said about a question he never addressed. It generates something plausible, in a plausible voice, based on patterns in his writing. Most of the time that output will sound reasonable. Sometimes it will be subtly wrong, or confidently wrong, and there is no way for the person reading it to tell the difference unless they already know the source material well enough not to need the tool in the first place.

It is also fair for some people to feel uneasy about the idea itself, however carefully it is labeled. Putting words in the mouth of a scholar who spent a lifetime earning the authority to speak on these matters is not something to do casually, and a disclosure banner does not fully resolve that discomfort on its own. That unease is worth taking seriously, not designing around.

Why the label has to be explicit

This is why any feature of this kind needs a clear, unavoidable label: this is an educational simulation, not the scholar's actual words. Not a footnote. Not something buried in a settings page. Visible at the moment someone reads the answer.

The reason is simple. Religious authority carries weight that a product feature does not. If someone reads a fabricated interpretation and believes it came, in some meaningful sense, from Al-Ghazali himself, that belief can shape how they think about their deen. That is a different category of harm than a wrong answer about, say, a restaurant recommendation.

A wrong restaurant recommendation costs you a bad meal. A fabricated fatwa costs something else entirely.

The safer version of the feature

A well-built version of this idea still has real value. It can introduce people to a scholar's actual body of work, point to real texts, and make dense classical writing approachable. Done honestly, it should:

• State clearly, every time, that responses are a simulation for learning, not the scholar's real words or a religious ruling

• Ground responses in that scholar's actual writing rather than inventing positions from nothing

• Say plainly when a question falls outside what the scholar addressed, instead of generating a confident guess

• Point people toward real primary sources and qualified scholars for anything serious or personal

Show your work, not just a disclaimer

A single disclosure line is necessary but not sufficient by itself. A more honest design goes further and shows how grounded a specific answer actually is: whether it came from a direct hit in a real retrieved source, a reasoned inference drawn between sources, or genuinely falls outside what the material covers. Treating every answer as equally confident, when some are load-bearing citations and others are educated guesses, quietly undoes the value of the disclosure sitting above it.

The same honesty applies to the review process behind the feature, not just the output. If scholarly review of the underlying content is still a work in progress rather than complete, that is worth saying plainly rather than implying a level of vetting that is not there yet. Admitting a gap is still open is a form of the same trust the disclosure is trying to build.

Convenience is not the same as authority

It is worth being honest about the tradeoff here. A simulation is more approachable than a 900 page classical text. That approachability is genuinely useful for building interest and understanding. But approachable is not the same as authoritative, and a product that blurs that line, even with good intentions, is trading a user's trust for a smoother experience.

The test is simple: would a reader still understand the limits of what they just read if the disclosure disappeared tomorrow? If the answer is no, the feature is leaning on the label to do work the design itself should be doing.