AI gives everyone the same answer.
The method gives you intelligence.
Ask ChatGPT "who are my competitors in Padova?" and you'll get a list of names with their Google stars. All accurate, sure. But your competitor asked the exact same question and got the exact same list. The restaurant across the street, the firm next door, the shop at the end of the block. Everyone looking at the same names. Where's your edge? Nowhere.
This guide teaches you a different method. It's called OSINT, Open Source Intelligence, or more precisely, it's inspired by the OSINT method. Investigators and journalists use it to find what isn't on the surface.
Open Source Intelligence (OSINT) is a structured method for the planning, collection, processing, analysis and dissemination (the classic "OSINT Cycle") of public data to create a competitive edge. Unlike simple keyword-based searches, business OSINT pulls out intelligence you can act on by cross-referencing unconventional sources like government grants, public registries and "Grey Literature", while filtering out the background noise.
Grey Literature covers all the high-value strategic documents produced by companies, governments or academia outside traditional commercial publishing channels. Think technical white papers, university theses, unpublished reports and investor presentations. It's a real gold mine for OSINT, because it hides raw data, technical specs and market trends long before they become public knowledge.
We adapt it to your business using free AI tools.
DATA vs INTELLIGENCE
| What AI gives you today | What you get with the method |
|---|---|
| A list of competitors | Who's growing and who's closing |
| Google stars | Why the stars are dropping (and what that tells you about the market) |
| Restaurant X is highly rated | Restaurant X has been hiring for 4 months (likely internal trouble) |
| Names and addresses | There's a 50,000 euro grant and none of your competitors know about it |
The problem isn't that AI gets things wrong. The problem is that it gives you data. And everyone has data. What you need is intelligence. Information your competitors don't have. Connections they don't see. Opportunities they don't know exist.
The method in 5 phases
BRIEFING. The question nobody asks. Not "who are my competitors" but a structured assignment that pulls out real signals.
COLLECTION. Perplexity with verifiable sources. The starting point, not the finish line.
DIGGING. The chain (ask "why?"), triangulation (compare 2 AIs), hidden sources (grants, institutional data).
DOSSIER. NotebookLM extracts without making things up. Gemini reasons over the clean data and produces the Gap Analysis.
DECISION. Act on intelligence, not on data your competitors have too.
The 3 levels
You don't have to use everything every time. The right level depends on the decision.
| Level | What you do | Time | When to use it |
|---|---|---|---|
| BASIC (Level 1) | Structured briefing + Perplexity with sources | 5-10 min | Everyday decisions. A market data point or a quick trend |
| ADVANCED (Level 2) | Digging (chain + triangulation + hidden sources) + Dossier (NotebookLM + Gemini + Deep Research) | 30 min | Important decisions. Market analysis or evaluating an investment |
| AGENT (Level 3) | Manus / Operator run the whole method for you. You supervise using the method. | 5 min + supervision | Strategic decisions. A new market or a partnership |
PHASE 1. How to structure an effective OSINT briefing
AI gives everyone the same answer because everyone asks the same question. Your edge isn't the tool. It's the question.
An investigator doesn't ask "what's out there?". They ask "who opened in the last 6 months, who's closing, and why?". Different question, different information. Same AI, a completely different result.
The briefing below isn't a question. It's an assignment for an OSINT analyst. It tells the AI to look for specific signals. Job postings, which show whether a company is growing or in trouble. Reviews, for what customers really think. Institutional data like grants, funding and permits. Public documents like price lists and catalogs. Every section of the briefing is there for a reason.
Done
Copy the template, replace [INDUSTRY] and [CITY/REGION] with your own, and paste it into Perplexity or ChatGPT. Then compare the result with "Who are my competitors in [city]?". You'll see the difference right away.
How to qualify a company before you call it (15 minutes)
The method isn't just for studying competitors. It also works before a sales call: five steps, public sources only, no paid tools. At the end you know whether that company is worth the call or not.
Who they really are. Where to look: the Business Register filing (visura), filed financial statements, the National State Aid Register (RNA). What to note: year of incorporation, legal form, latest available revenue and which way it's trending, public grants received.
Whether they're growing or struggling. Where to look: it.linkedin.com/jobs, Indeed, InfoJobs, over the last 6 months. What to note: how many open positions and what kind. Key roles and new functions mean expansion. The same three entry-level jobs reposted every two months mean turnover: a problem nobody has solved.
What their customers say. Where to look: Google Maps, Trustpilot, industry portals. What to note: average rating, how many reviews in the last 6 months and, above all, the three complaints that keep coming up. Those are the words you'll use on the call.
What they're already doing. Where to look: their website, blog, social pages, and the Wayback Machine to compare with a year ago. What to note: date of the last update, suppliers and tools you can see, the promises they make in public.
What they didn't put on the homepage. Where to look:
site:it "Nome Azienda" (listino OR catalogo OR bilancio) filetype:pdfand the town's official notice board (Albo Pretorio). What to note: prices, permits, work in progress, expansions.
How to decide. The prospect is qualified if the five steps give you three things: a problem you know how to solve and that you read somewhere, not imagined; a signal that they can spend right now (they're hiring, they won a grant, they just opened); the name of the person who decides, not the info@ address.
No documented problem means it's not a prospect, it's a name on a list. No spending signal means the company goes on your calendar for three months from now, not on today's call list. No name means you're handing the call to the switchboard.
Info
Sign up to get the complete OSINT workflow with the prompts for collection and digging, the triangulation techniques, the Extractor + Referee workflow for the dossier, AI agents and the decision checklist.
LEVEL 1 — BASIC How to collect the data: Perplexity with sources
Perplexity and ChatGPT both cite sources in their answers now. The difference is elsewhere. Perplexity runs a specialized search algorithm built to dig deeper into web sources. The paid version (Pro Search) asks clarifying questions before searching and returns much more targeted results. On top of that, Spaces let you organize your research into persistent workspaces and query them over time, like a dossier that keeps growing.
For the collection phase, Perplexity is the recommended starting point. It's free, and you don't even need an account.
Open Perplexity and paste in the OSINT briefing you copied in the previous section, with your industry and city filled in. That's your collection prompt. You don't need any other prompts for this phase.
LEVEL 2 — INTERMEDIATE How to dig below the surface: 3 essential techniques
The first result is the floor, the stuff everyone has. The value is underneath. An investigator never stops at the first source. They cross-check. They triangulate. They look for confirmations and contradictions.
Three techniques, ten minutes total. Your competitor isn't doing this.
Technique 1. The chain (the "why" rule)
When AI gives you a data point, like "the industry is down 3%", almost nobody ever asks why. Yet it's the most powerful question you can ask. The first result is the headline. The "why" is the article. And the article is where the value is.
Three chained questions take you from "a list of restaurants" to "understanding who's winning and why".
Info
How it works
Run the briefing on Perplexity.
On the most interesting data point, ask "WHY?"
Example: "Why are restaurants closing in Padova? What are the specific local factors, not just the national ones?"On the exception, ask "WHO'S THE EXCEPTION?"
Example: "Are there restaurants growing despite the downturn? What are they doing differently?"
3 questions = 3 levels of depth. The end result isn't a list anymore, it's an analysis.
Technique 2. Triangulation to cut hallucinations by 80%
No business owner accepts the first quote. But with AI, everyone accepts the first answer. Triangulation is simple. Ask two different AIs the same question and compare.
Info
How it works
Ask Perplexity and ChatGPT the SAME question. Then compare.
If they match, the data point is solid
If they don't match, you've found something worth looking into
If one finds things the other doesn't, use them together
This isn't about which AI is "better". It's about using them together.
Info
When you find data that doesn't match between the two AIs, don't stop to figure out which one is right. Save the pages as PDFs. At this level your job is to collect the mess, not untangle it. The AI does the untangling in the next step, in two precise passes.
Technique 3. Hidden sources
This is the most powerful technique in the guide. Ask the AI to find institutional sources you'd never have checked on your own. It finds real money. Grants, funding, free programs.
Heads up
This prompt finds real money. Chamber of Commerce grants, non-repayable funding for going digital, free innovation programs. Thirty seconds for a question that could be worth thousands of euros to your business. AI would never have given you this with "who are my competitors".
Total digging time: 10 minutes. With three extra questions and two AIs, you have a market analysis with trends, exceptions and grants. Your competitor who asked the generic question has none of this.
QUICK REFERENCE Regular prompt vs Deep Research
Every AI tool has two modes. The regular prompt, where you ask a question and get an answer in a few seconds. And deep research, where you hand over an assignment and the tool reads dozens of pages and produces a report for you. It's the difference between asking for information and commissioning an investigation.
| Tool | Regular prompt | Deep Research (how to turn it on) |
|---|---|---|
| Perplexity | Type and send. Answer in 10-30 sec. | Pro Search toggle at the top. It asks you questions, then searches in depth. 1-3 min. FREE (5/day) |
| ChatGPT | Type and send. Answer in 10-30 sec. | Model menu at the bottom left > Deep Research. Reads 50+ pages and produces a report. 5-15 min. ChatGPT Plus ($20/month) |
| Gemini | Type and send. Answer in 10-30 sec. | Deep Research icon in the toolbar. Reads 30+ pages and produces a report. 3-10 min. FREE |
Info
Start Deep Research in the background (it takes a few minutes) and meanwhile work with regular prompts in the other tools. When it's done, you have a report to add to the dossier.
LEVEL 2 — ADVANCED Deep Research: the right prompt
For Deep Research, the prompt has to be an assignment, not a question. You're commissioning an investigation. Deep Research reads 30-50 web pages and produces a full report. It takes 5-15 minutes. Start it, and build the dossier with NotebookLM and Gemini while you wait.
LEVEL 2 — ADVANCED How to build a bulletproof dossier with NotebookLM (the Extractor)
An investigator doesn't keep the evidence in their head. They put it in a dossier. A place where they can find everything again, connect the pieces and ask it questions.
If you take the research PDFs and dump them straight into Gemini saying "analyze everything", Gemini hallucinates. It fills the gaps with plausible data and builds a coherent story. One that's partly made up. It's not its fault, it simply doesn't know what it doesn't know.
That's why we work in two steps. First the Extractor, which maps without making things up. Then the Referee, which reasons over the clean data.
Info
How to use NotebookLM
Go to notebooklm.google.com (free, all you need is a Google account)
Click "New notebook"
Click "Add source" and upload the PDFs from your Perplexity and ChatGPT research and the grants you found. Different sources, different data, sometimes contradictory. Raw chaos.
Wait for it to process (10-30 seconds)
In the chat, paste the prompt below. Don't ask "give me an analysis". Give it a rigid template.
Done
NotebookLM answers ONLY with what's in your documents. The CONTRADICTIONS section is the one that matters most. It shows you where the sources disagree, without you having to read them all. Where data is missing, it writes "DATA NOT PRESENT". That's valuable information too. Copy the list it produces and move on to the next step.
LEVEL 2 — ADVANCED How to get a strategic Gap Analysis from AI (the Referee)
The Referee takes the data NotebookLM has already filtered and produces the Gap Analysis Matrix.
Gemini doesn't get the raw PDFs, it gets data that's already clean. That's why it doesn't hallucinate. It has no gaps to fill, so it can reason, settle conflicts and produce concrete actions.
Info
How to do it (5 minutes)
Copy the bulleted list NotebookLM produced (the previous step)
Open gemini.google.com (or claude.ai), free with a Google account
Paste the NotebookLM list first, then the prompt below
Look at the table it produces. Gaps, causes, difficulty, and what to do tomorrow morning
You can do this in 5 minutes. A consultant would take a day and charge you 500 euros. You have two free tools working in sequence. The Extractor cleans up, the Referee reasons.
| Role | NotebookLM (Extractor) | Gemini / Claude (Referee) |
|---|---|---|
| Input | Upload the raw documents | Gets data that's already clean |
| Behavior | Extracts ONLY what's there | Reasons, referees, decides |
| Superpowers | Maps the contradictions | Finds the gaps and the actions |
| If data is missing | Writes "DATA NOT PRESENT" | Builds the Gap Analysis Matrix |
| In the workflow | Workflow input | Workflow output |
LEVEL 2 — ADVANCED Audio Overview (NotebookLM)
NotebookLM has a feature that turns your documents into a podcast. Two voices discussing your market research. A regional grant becomes a spoken explainer. It feels like magic, but it's just a smart use of your data.
Info
How to do it
In the notebook, click "Audio Overview" (the audio icon at the top)
Click "Generate" and wait 2-5 minutes
Hit play and listen
Handy in the car on the way to work. For walking a partner through a document without making them read it. For a quick refresher before a meeting. For getting through a grant or a contract without reading it word for word.
LEVEL 3 — AGENT AI agents
Everything you've done so far, from the briefing to the dossier, an AI agent does in five minutes.
An agent isn't a chatbot. It doesn't wait for your questions. You give it an assignment, exactly the kind of briefing you've learned to write, and it sets off on its own. It opens sites, reads pages, compares data, and when it's done it hands you a report.
| Feature | Chatbot | AI agent |
|---|---|---|
| How it works | Answers a question and waits for the next one | Gets an assignment and works on its own |
| Autonomy | Sits still until you talk to it | Opens sites, reads pages, compares data, writes reports |
| Analogy | Like an employee waiting for orders | Like an investigator you hand a case to |
| Agent | What it does | Cost |
|---|---|---|
| Manus (manus.im) | Takes an assignment, browses the web, collects data, produces a full report with sources | Free (with limits) |
| ChatGPT Operator | Browses websites for you, fills in forms, compares products. Like having a digital intern. | ChatGPT Plus ($20/month) |
| Claude Computer Use | Takes control of your computer: opens the browser, navigates, reads pages, fills things in. You watch. | Claude Pro ($20/month) |
| Perplexity Spaces | Workspaces where you build up research over time. A dossier that grows every time you search for something. | Free (basic) |
The agent prompt is the same briefing from the method, expanded with the digging instructions. The agent runs the whole method for you.
Heads up
These agents work well 70-80% of the time. Sometimes they get it wrong, sometimes they get stuck, sometimes they give you partial results. That's why you learned the method before the agents. You know how to check the result. You know how to run the chain, how to triangulate, how to find hidden sources.
The method is your compass. The agent is the autopilot. But you're always the one at the wheel.
PHASE 5: Decision
Now you have intelligence, not data. The only difference between people who use AI like everyone else and people with a competitive edge is the method. Before you act, verify.
Done
Checklist before you decide
Do the different sources agree on the key points?
Did the digging confirm or contradict the first search?
Does the dossier cover every aspect of my briefing?
Are there hidden opportunities (grants, trends, niches)?
Are there unresolved red flags?
If yes to 1-4 and no to 5: you can decide with confidence.
If something doesn't add up: go back to Phase 3 and dig deeper.
Which level to use when
Info
EVERYDAY DECISION (5-10 minutes)
"I need a market data point" / "I need to understand a trend"
→ Perplexity + structured briefing → LEVEL 1
IMPORTANT DECISION (30 minutes)
"I need to analyze a market" / "I'm evaluating an investment"
→ Digging (chain + triangulation + hidden sources) + Dossier (NotebookLM + Gemini + Deep Research)
→ LEVEL 2
STRATEGIC DECISION (5 min + supervision)
"New market" / "Partnership" / "Expansion"
→ Agent (Manus / Operator) + supervision using the method
→ LEVEL 3
Never skip the basics. Even when you use the agent, you need the method to check the result. Tools change, the method doesn't. Briefing, collection, digging, dossier, decision. This pattern worked for investigators in 2005 and it will still work in 2030.
Links and resources
| Tool | Link | Level | Free? |
|---|---|---|---|
| Perplexity | perplexity.ai | Level 1 | Yes (no account) |
| ChatGPT | chat.openai.com | Level 1-2 | Yes (basic) |
| ChatGPT Deep Research | chat.openai.com | Level 2 | ChatGPT Plus ($20/month) |
| Gemini | gemini.google.com | Level 1-2 | Yes (Google account) |
| NotebookLM | notebooklm.google.com | Level 2 | Yes (Google account) |
| Manus | manus.im | Level 3 | Yes (with limits) |
| ChatGPT Operator | chat.openai.com | Level 3 | ChatGPT Plus |
| Perplexity Spaces | perplexity.ai | 2-3 | Yes (basic) |
| Resource | Link | What you'll find |
|---|---|---|
| Chamber of Commerce | camcom.gov.it | Local grants, regional economic data |
| Regional government (varies) | regione.[nome].it | Regional grants, SMB incentives |
| Confcommercio | confcommercio.it | Programs, training, association grants |
| MISE | mise.gov.it | National incentives for businesses |
| Invitalia | invitalia.it | Subsidies and funding for startups/SMBs |
| National State Aid Register | rna.gov.it | Public funding received by competitors |
| Google Fact Check | toolbox.google.com/factcheck/explorer | Claims that have already been fact-checked |
Info
Tomorrow morning, open Perplexity and run ONE real search for your work. Not a test, a search you actually need. With the briefing. Two minutes. If you do it tomorrow, it becomes a habit. If you don't, it stays a guide.
Cosa ottieni iscrivendoti:
- Accesso a tutti gli episodi della newsletter
- Guide e corsi completi sull'AI per marketer
- Strumenti AI professionali (BrandPix, Short Video Suite)
- Crediti gratuiti per iniziare subito