Mauro Giorgi
I conduct due diligence on AI-generated documents:
I detect, verify and correct artificial-intelligence hallucinations.
On 22 August 2026, I asked artificial intelligence for four market analyses.They were false.
That day I spent the entire morning building a market analysis with one of the most advanced language models available.
It delivered four documents: tables, percentages, competitor comparisons and revenue projections. They were well written and appeared rigorous. I could have used them to decide where to invest for the next two years.
They contained false statements.
Not fabricated citations — those are found by looking for a source and failing to locate it. These were harder to see: a market described as uncontested when it was not; a company said not to serve a segment it had served for years; numbers calculated precisely from costs no one had verified.
I found them because I kept asking for verification — four times, against an answer that already looked complete. Had I stopped at the first response, I would have built a project on a market that did not exist.
Four false statements.What was true
“No competitor serves this market.”
“That provider serves banks and large organisations. Small businesses are excluded.”
“No one sells predictive recommendations to small businesses.”
“110 customers for €16,190 per month.”
Every crossed-out line was plausible, consistent with the rest of the document and better written than I would have written it. That is what makes it dangerous.
Why I decided to work on this
This is not an isolated incident. A language model produces unsupported statements whenever no one asks it to substantiate them — and it will continue to do so, because this is a property of how it works, not a defect to be fixed in a future release.
The point is that responsibility remains with the signatory: it does not pass to AIwithMauroGiorgi. Italian law states this expressly for intellectual professionals — Article 13 of Law No. 132 of 23 September 2025 provides that artificial intelligence may serve only as an instrumental aid and that the intellectual work forming the subject of the service must remain predominant.
The Italian Supreme Court, Criminal Section III, judgment No. 34481 of 22 October 2025, said this in relation to judges, but in terms that apply to anyone signing a document. Digital tools make drafting easier — and for that very reason increase the risk, the Court wrote, of taking the grounds for a decision aliunde, “abdicating the duty to contribute one’s indispensable and irreplaceable evaluative judgment”.
Indispensable and irreplaceable: no one can evaluate in your place, and no one bears responsibility in your place.
The practical consequence concerns insurance cover. Those who systematically verify output remain within ordinary negligence, the area professional indemnity policies typically cover. Those who delegate without control expose themselves to gross negligence, where cover may not respond — and some policies already contain express exclusions for losses caused by IT malfunctions or third-party software defects.
Another point removes every excuse: the fact that these systems produce errors is now common knowledge among reasonably informed professionals. Ignoring it is no longer naïveté.
Having recognised all this, I decided to build AIwithMauroGiorgi to review AI output systematically. I built it first of all for myself, because I am responsible for what I approve and sign in my name. Then I realised it was useful to anyone in the same position: professionals, business owners and anyone signing a document on which someone else will make a decision.
Verification is possible. But it must be done every time, on every document, and leave a record — otherwise it remains a good intention that no one can prove.
About me
I spent almost thirty years on the side of people risking their own money: first with a communications agency working for international fashion and accessories brands around the world. I later lived abroad — including time on both the East and West Coasts of the United States — before settling for more than ten years in Saint-Barthélemy, where I founded Carla St Barth, a luxury multi-brand boutique in Gustavia that remains among the island’s best. I do not come from artificial-intelligence research. I come from places where decisions carry a cost and problems must be stopped at the outset.
Today I build tools with AI — and verify what AI produces.
Receive AIwithMauroGiorgi.It is free
I have sent you an email with the AIwithMauroGiorgi link and instructions.
If it does not arrive within a few minutes, check your spam folder.