Briefs and inboxes are drowning in mass-generated proposals. Here are the concrete signals that separate an AI consultant who has shipped to production from a seller of slides.
The pitch arrives polished: confident claims about "leveraging AI to unlock synergies", a list of buzzwords that could apply to any industry, and a promise that a chatbot will transform your business in two weeks. It reads well. It also reads like it was written by a language model in ninety seconds, because it probably was.
This is AI slop: high-volume, low-substance output generated to win attention rather than solve problems. It has spread from blog comments into procurement inboxes, and it makes hiring genuinely harder. The good news is that slop leaves fingerprints. Here is how to spot them, and what to look for instead.
The tells of a generated proposal
Mass-produced AI pitches share recognizable patterns. None is damning alone; three or four together usually are.
- Generic specifics. The proposal praises "your innovative company" without naming your sector, your stack, or a single constraint you mentioned. Real consultants mirror your words back because they listened.
- Effortless confidence, zero trade-offs. Every project is feasible, fast, and cheap. Experienced builders talk about limits: data quality, latency, cost per query, compliance.
- Impossible timelines. "Production-ready RAG assistant in five days" ignores evaluation, guardrails, and integration work. Anyone who has shipped knows the demo is ten percent of the job.
- No questions. A serious consultant asks about your data, your users, and what happens when the model is wrong. Slop never asks, because it is not listening.
- Buzzword density. "Synergistic AI-driven transformation" says nothing. "A retrieval pipeline over your internal documentation, with citations and a fallback to human support" says something testable.
Five questions that separate builders from sellers
Use these in the first call. The answers matter less than whether the person can answer them concretely.
1. "Walk me through a system you built that is still running."
Ask for architecture, not adjectives. Which model, hosted where, at what cost per thousand requests? How is it monitored? What broke in month two? A builder answers in specifics and volunteers problems. A seller pivots back to capabilities.
2. "What would make you tell me not to do this project?"
Honest consultants turn work down. If AI is the wrong tool for your problem, they should say so. Watch for a candidate who names a cheaper non-AI solution: that is a strong signal of integrity.
3. "How will we know it works?"
Push for evaluation before build. What does success look like: accuracy on a labeled sample, deflection rate, hours saved per week? Anyone proposing an AI project without a measurement plan is selling a demo, not a system.
4. "What happens when the model is wrong?"
Every AI system fails sometimes. Good answers involve confidence thresholds, human review, logging, and graceful degradation. Vague answers about "continuous learning" mean the question has never been considered.
5. "Who owns the code and the prompts?"
Ask about handover, documentation, and what happens if you stop working together. Lock-in through obscurity is a business model; clarity is a service.
What good engagement looks like
A trustworthy freelance AI consultant will typically start small: a paid discovery or a scoped pilot with a defined deliverable and an exit. They will ask for sample data before quoting. They will separate the prototype from the production system in both timeline and budget, because those are different engineering problems. They will mention costs you did not ask about: inference bills, evaluation effort, maintenance.
If you are evaluating an AI project, treat the proposal itself as a work sample. Is it specific to you? Does it contain a number you can verify? Does it admit uncertainty anywhere? Slop is frictionless; real expertise has edges.
A practical filter you can apply today
Before your next call, send one question: "Describe the last AI feature you shipped that disappointed you, and why." Sellers will answer with a success story. Builders will tell you about the retrieval system that failed on scanned PDFs, or the classifier that drifted when the input distribution changed. That answer, more than any portfolio, tells you who you are talking to.
Whether you write in English, French, or Arabic, the same signals hold: specificity, trade-offs, and a willingness to say "it depends, and here is what it depends on." Those qualities do not translate poorly, and they cannot be mass-generated.
Let's talk about your project
If you are weighing an AI idea and want a straight assessment, I am happy to discuss it. Bring your constraints and your skepticism; the first conversation is about understanding your need, not selling you a model. You can reach me at contact@hamzabelgacem.com.