Why AI Analysis of Your Ad Accounts Needs Verification
Analyses of ad accounts made with Claude, Gemini or ChatGPT need to be checked by a specialist, because they are written very well even when they are wrong. Giorgio Terenziani of ERMETHE describes a case in which 80% of an analysis did not hold up against the strategy, and explains where AI really works.
Video in Italian with English subtitles.
Key takeaways
- AI writes coherent, convincing analyses even when it has not understood the strategy: good writing is not a guarantee.
- Giving context once is not enough: you have to keep asking questions, verifying and falsifying the data.
- In the example shared in the video, 80% of an analysis made with Claude was not true when compared with a specialist.
- AI works very well on mass-scale tasks; for strategic choices you need to talk it through with a person.
- A phone call with the specialist is often worth more than a long generated report, especially in periods like Black Friday.
Learn more: MCP and performance marketing: the full article
Watch it on Instagram
Video transcript (translated from Italian)
What a struggle. Okay, let's clear something up, let's talk it over.
Claude, Gemini, ChatGPT are really good at doing analyses on your accounts too. We use them, we love them a lot. People talk a lot about context, but also the ability to ask questions again, not only giving the context once, but continuing to ask questions, verifying, falsifying the data that these very powerful tools give us, is fundamental.
Why? Because a response comes out that is wonderful in terms of writing, that seems perfectly coherent. Then you compare it with a person, a specialist (luckily at ERMETHE we have specialists, which is really nice), and you see that 80% of what Claude told you, to give an example, doesn't work, isn't true, it didn't understand the macro strategy, even though it had the context. Where does it really work well? On mass-scale tasks.
So it's great to do analyses. But why did I start by sighing? Because often on Monday morning, like today, or sometimes on Tuesday too, we find ourselves answering questions from clients, who rightly use these tools to do analyses, and even more in the middle of Black Friday. But at the same time, these are questions that take time away from the team to do things that would be more useful, because they're poorly framed.
So it's better to talk it through, better a phone call. Yes, the good old telephone, this one here. This one is old, but it was to give an example. The good old telephone, to figure out together what to do, because they work great, but you have to ask the right questions.
Original transcript in Italian
Che super fatica. Allora, chiariamo una cosa, parliamo insieme.
Claude, Gemini, ChatGPT sono bravissime a fare analisi anche sui vostri account. Noi li usiamo, li amiamo tantissimo. Si parla tanto di contesto, ma anche la capacità di riproporre domande, non solo dare il contesto una volta, ma continuare a fare domande, andare a verificare, falsificare i dati che ci danno questi strumenti potentissimi, è fondamentale.
Perché? Perché esce fuori una risposta meravigliosa a livello di scrittura, che sembra perfettamente coerente. Poi tu la vai a confrontare con una persona, uno specialist (per fortuna noi in ERMETHE abbiamo specialist, questo è veramente bello), e vedi che l'80% di quello che ti ha detto Claude, per fare un esempio, non funziona, non è vero, non ha capito la strategia macro, anche se aveva il contesto. Dove funziona veramente bene? Sulle cose di massa.
Quindi benissimo fare analisi. Ma perché ho cominciato sospirando? Perché spesso il lunedì mattina, come oggi, o delle volte anche il martedì, ci troviamo a rispondere alle domande dei clienti, che giustamente utilizzano questi tool per fare analisi, e ancora di più in mezzo al Black Friday. Ma allo stesso tempo sono domande che portano via tempo al team per fare cose che sarebbero più utili, perché sono fatte male.
Quindi meglio un confronto, meglio una telefonata. Sì, il buon vecchio caro telefono, questo qui. Questo è vecchio, ma era per fare un esempio. Buon vecchio caro telefono per capire insieme cosa fare, perché funzionano benissimo, ma bisogna fare le domande giuste.


