AI in Cybercrime Investigation: How Police Can Use Meta AI for Free OSINT
Investigators normally pay for scraping tools. Meta AI reads public Facebook, Instagram and Threads posts for free, with source links. Here is how police can use it for OSINT, with example prompts.
AI has become a buzzword across the world, and technology companies were among the first to adopt the shift. The growing use of AI to commit cybercrime has become a real concern for law enforcement agencies everywhere. But there is a silver lining: if investigators turn to AI too, it can be an immense addition to their capability. This guide shows how police officers, without any deep knowledge of large language models, can use AI in day-to-day investigation and open-source intelligence (OSINT).
- Meta AI can read and summarise public posts across Facebook, Instagram and Threads.
- Investigators normally pay for data-scraping and analytics tools. Meta AI does much of this for free.
- Sign in at meta.ai (you can use a Facebook login), then use clear, well-framed prompts.
- Treat every result as a lead to verify, not proof. This is an aid to investigation, not a substitute for it.
Why Meta AI for social-media OSINT
Meta has launched its own AI assistant, and it has an excellent ability to read through public content on Facebook, Threads and Instagram. To research public posts and groups, investigators usually spend a lot of money buying data-scraping and analytics tools. Meta AI makes much of that research easier, and most importantly, it is free. It also returns a hyperlink to each post it finds, pulled from Facebook or Instagram in any language, so an officer can go straight to the source.
Step 1: Create an account
Go to meta.ai and create an account. You can sign in using your existing Facebook account. No special access or paid subscription is needed to begin.
Step 2: Write a good prompt
The quality of the result depends almost entirely on the prompt. To make the AI understand your requirement, set the context first, state exactly what you want, and ask for the output in a structured form (for example, a table with links). The examples below show how a well-framed prompt turns a vague question into an investigation-ready list.
Example 1: Monitoring scams across regions
A prompt asking Meta AI to surface recent scam warnings and awareness posts, by date and country, returns a structured picture of what is being reported where. A sample of the kind of output it produces:
| Date | Region | Scam type | Source | Summary |
|---|---|---|---|---|
| Jul 4 | India | Fake job / investment ads | Cyber Dost (I4C) | Scammers buying FB/IG/Google ads targeting 18–34 year-olds; directs to helpline 1930 |
| Jun 29 | Philippines / Global | Account-hijack phishing | Feeling bryt | Fake "monetization violation" notice with a surge.sh link to steal credentials |
| Jun 29 | United Kingdom | Travel / holiday scam | Mumslifemcr | £400 caravan booking via a FB ad with fake reviews; the caravan did not exist |
| Jun 29 | United Kingdom | Financial deepfake ads | Martin Lewis | Warns that social ads should be assumed to be scams; his image used in death-hoax ads |
| Jun 30 | United States / Global | "Fake death" phishing | Jeff Rossen | Tagged death notices link to a fake FB login to harvest passwords |
| Jun 30 | Germany | Meta-impersonation phishing | Mimikama | Scammers buy FB ads pretending to be Facebook, leading to a fake Windows lock screen |
| Jun 30 | New Zealand | Employment scams | New Zealand Police | Red flags: unsolicited offers, moving chat to WhatsApp/Telegram, app downloads |
| Jul 1 | Nepal | E-commerce bait-and-switch | Aditya Gurung | ₹499 gimbal ad delivered a ₹1,499 COD selfie stick |
| Jul 1 | United States | Creator identity theft | Howry Family | A fake LLC created in their names to siphon Facebook video revenue; a follower got a fraudulent 1099 |
| Jul 2 | Zambia | Fake recruitment ads | Young Phiroz General Dealers | Job ad for cashiers/cleaners marked FAKE; applicants asked for a K300 uniform fee |
Example 2: Finding accounts promoting restricted content
You can also ask Meta AI to list public profiles of a particular nature. Set the context first, then give the prompt. Tobacco promotion is used here only as a reference point, because its advertising is restricted in India under local regulations, which makes it a clean test case for the technique.
Note: the results below are examples the AI surfaced for an investigator to review. They are not a legal finding, and no wrongdoing is established by their inclusion here.
Prompt: "Create a table. Give only the links and a short description of what is in the video, for accounts promoting tobacco, including in regional languages."
| Link | What the video shows |
|---|---|
| reel/1432925565110002 | Close-up of a large dark compressed bundle of khaini leaves in an open-air yard in Vaishali, Bihar; a folk jingle plays; comments quote ₹500–600/kg. |
| reel/1659039405471046 | Selfie video inside a small shop in Gurdaspur; the seller promotes his khaini and says he ships to Delhi, Mumbai and abroad. |
| reel/1411193556977702 | A man speaks in Hindi, introduces a relative in Hyderabad who prepares khaini, and invites viewers to contact him for supply. |
| reel/1326256369340261 | A seller stands in front of shop shelves, advertises a khaini shop on NH31, lists varieties and shows a phone number for orders. |
| reel/1428727402133726 | Business promo for a khaini store in Ward 5; a bright signboard, a chaff machine, and khaini, churot and paan masala on display. |
| reel/1708946213465547 | Warehouse tour in Bihiya, Bihar; the speaker points to rows of jute sacks and an old processor and pitches khaini at ₹500/kg. |
| reel/1360359032709087 | A vendor at a market chowk; sorting tables and packets being packed, with landmark directions to the stall. |
| reel/1766197144735231 | A roadside kiosk near Bengal Gate; the seller holds a small red-string bundle and says khaini is available for ₹50. |
| reel/977133972152063 | Close-ups of a purple-labelled container and a green cylinder pack of branded khaini inside a mobile shop. |
| reel/1231911592467255 | Indoor ad for a Nepali khaini brand; red foil packets displayed on a counter, with viewers asking for the price in comments. |
Source: Facebook reels surfaced for khaini promotion, January–July 2026.
Step 4: Sentiment and the emotional palette of a region
The same approach extends to mood. You can ask Meta AI to summarise how a particular topic, event, scam or campaign is being discussed within a region, and to break down the emotional palette behind that conversation. This lets an officer gauge public reaction, catch rising panic or anger early, and track how a narrative is spreading, all without manually reading thousands of posts.
Prompt: "For public posts in [region] discussing [topic] over the last two weeks, give me: (1) the overall sentiment split, (2) the dominant emotions and roughly how common each one is, and (3) three representative posts with links, in any language."
A sample of the kind of structured output it returns, for a viral loan-app scam in one state, is shown below. The figures are illustrative of the format, not fixed values.
| Measure | Reading |
|---|---|
| Overall sentiment | Negative 68% · Neutral 22% · Positive 10% |
| Dominant emotions | Fear (most common) · Anger · Distrust of authorities · Shame (under-reported) · Hope (recovery / helpline posts) |
| Rising signal | A sharp jump in fear and anger in the last 3 days, clustered around one district |
| Representative posts | Three linked posts: a victim describing harassment, a warning shared by a local group, and a helpline (1930) reminder |
Because each result carries a hyperlink to the underlying Facebook or Instagram post, an officer can move straight from the summary to the source and confirm the reading. It also surfaces the emotional tone that raw numbers miss, for example widespread shame that keeps victims from reporting, which can shape how a public advisory is worded. Treat sentiment scores as an indicator, not evidence: a spike is a reason to look closer, not a conclusion.
Limitations and cautions
- Verify every result. AI can make mistakes, miss content, or misread context. Treat each link as a lead to check, not a conclusion.
- Public data only. This technique reads public posts. It is not a substitute for a lawful data request to the platform when you need non-public account records.
- Stay within your authority. Use it only for authorised investigations and in line with your local law.
Closing thoughts
Meta AI is excellent at analysing content posted across its platforms, and it can be a genuinely useful, free aid for intelligence agencies and police officers around the world during investigation and analysis. Used with good prompts and a healthy dose of verification, it turns hours of manual searching into a structured, source-linked starting point.
For educational and training use by authorised investigators. AI output can be incomplete or incorrect and must be independently verified.