The OSINT Newsletter - Issue #91
The Investigator’s Best Friend: OSINT With AI
👋 Welcome to the 91st issue of The OSINT Newsletter. This issue contains OSINT news, community posts, tactics, techniques, and tools to help you become a better investigator. Here’s an overview of what’s in this issue:
Prompting: How to “Talk” to AI
The AI Team: Tools for AI OSINT
Example: OSINT with AI in Action
🏁 New CTF Challenge Live - The Missing CEO (3 Part)
A new CTF challenge has been posted on our CTF website. This weeks CTF was designed by @foilmanhacks and is very challenging.
Start competing in our Capture the Flag (CTF)
🪃 If you missed the last CTF, here’s a link to catch up.
Last weeks CTF challenge was an image analysis challenge called (Nothing to See Here) answer: Steghide was used to extract an embedded secret.txt from the JPEG (password found via rockyou.txt), the message pointed to the nearest campsite and warned about a Bosnia name decoy, and the terrain/coastline were matched on maps to pinpoint the viewpoint and identify the closest campsite as Campingplatz Vucine for the flag.
🪃 If you missed the last newsletter, here’s a link to catch up.
⚡ Why OSINT Certifications Aren’t Worth It and What to Do Instead
🎙️ If you prefer to listen, here’s a link to the podcast instead.
Let’s get started. ⬇️
OSINT data isn’t always user-friendly; it can be a hostile landscape of scattered sources, dense PDFs, super-long social threads and dodgy data leaks. Often, the main challenge is sorting out what’s actually helpful, and doing it fast enough to be useful.
That’s where AI comes in. At the moment, it seems like many people don’t love it- probably because they don’t do OSINT. If you’re an OSINT investigator, AI is your best friend… and in this issue, we’ll show you how to use it.
We’ll cover:
How to “talk” to AI
Essential AI tools (that actually work)
How to get accurate, repeatable results
And an example of AI in action
By the end, you’ll have met your new research assistant, and worked it into your workflow. Let’s get acquainted.
What Can AI Do For Me?
Well-deployed AI is extremely reliable and useful. It can:
Read and summarise more text than a human can
Draw out patterns or contradictions in data
Extract specific types of data
Cross-reference with other sources
Translate foreign languages (at a basic level)
Come up with fresh angles or unexpected pivots
In other words, AI lets you dump the time-consuming (and dull) parts of OSINT off onto a machine, leaving you free to focus on the important stuff. But of course, even a machine can’t do everything. AI can handle grunt work, but it won’t replace you or your typical OSINT tools. You still need the traditional tools to discover data, while AI helps you understand, present and use that data more effectively.
Prompting: How to “Talk” to AI
Although it seems like AI can speak any human tongue, it’s not the same as a person. You have to use its “language”: also known as prompting.
A prompt is a command that tells the AI what to do. AI follows instructions literally, so a good prompt gives good results. If you’re vague, the system will try to work out a solution by itself - usually with assumptions or just making things up. Think of it like a junior analyst who works incredibly quickly, but needs clear instructions so they don’t mess up. Or start hallucinating.
A good prompt usually follows this structure:
The role: who the AI should act like (e.g., “as an OSINT analyst”).
The task: what you want it to do (e.g., “extract all names from this document”).
The rules: what it should and shouldn’t do while completing the task (e.g., “cite sources,” “base all conclusions on the text”).
The output format: how you want the results delivered (e.g. “as a table”)
Why Do the Responses Still Suck?
If you’re clear about what you want, you’ll usually get it. But even if you think you’ve put in a good prompt, it might need a little refinement. Try these strategies.
Tell the AI Not to Guess: Add a line like: “If any information is missing or unclear, respond with ‘unknown’ rather than guessing.” This dramatically improves accuracy during investigations, and prevents hallucinating.
Break It Down: Instead of bombarding the AI with a string of complex instructions, break the task into little bits - each a separate query. Ask step-by-step: eg. 1. Extract detail, 2. Identify relationships, 3. Highlight inconsistencies - and so on.
Iterate: Work out what works for you on a tool-by-tool basis. If the output isn’t perfect, change your input. Ask follow-up questions. Provide examples. Try phrasing things differently… Then stick to whatever gets results.
The AI Team: Tools for AI OSINT
As you know, pro OSINT isn’t done with just one tool - it takes a whole team to make the dream work. Each one handles a different type of data: we’ve discussed AI image tools before, for example. We love it when an AI OSINT toolkit comes together, so try adding these AI-powered tool types to your loadout.
🤖 General-Purpose AI Analysts (ChatGPT, Claude)
These large-language models (LLMs) are like an extra brain to add to your investigation. On a basic level, they’re machine learning systems that work with text; so if you give them human language material (text or documents, for example), they’ll process it fast and return results.
They’re especially good at:
Summarising long docs with a succinct precis
Extracting data points like names and dates
Spotting suspicious details or unobvious inconsistencies
Creating strictly structured reports
Being the vanilla version of AI, these models are far from perfect. But with high quality prompts - and if they stick to what they’re good at - they’ll be extremely effective co-investigators.
🎨 Visualisers and Mapping Tools (Maltego, OSINT Industries Palette)
Visual investigation tools (like Maltego or OSINT Industries Palette) show how different pieces of information connect to each other within the context of your investigation - like a digital evidence board from a serial killer movie. And with the extra helping hand from AI, these tools get even stronger.
These tools can:
Visualise links between data points
Automatically add detail with AI-generated summaries
Get a clearer understanding of complex investigations
If you’re working on a case with lots of parties involved and are struggling to make links, these OSINT AI tools will straighten out the threads.
Archive and Capture Tools (Hunchly)
AI OSINT investigation tools don’t just help you understand what you’re investigating; they can also help you archive and present it. Apps like Hunchly will automatically record every page you visit during your investigation, taking all the drudgery out of archiving.
They can:
Collect the URL, timestamps, and hashes of every page you visit
Make full-page captures of pages
Categorize and tag captures (with search function)
Assemble findings into court-ready reports
They’re ideal for long investigations, and save you having to hunt through tabs or untitled files for that one screenshot you need. With OSINT AI tools, your investigation can be fully sourced and documented - no matter how much you browsed.
Large-Scale Data Processors (Elastic AI, Haystack)
Sometimes, OSINT work gives you huge quantities of data to sift through. From giant text dumps, to large leaks, and groups of thousands of documents, large-scale data processors allow you to blitz through tons of data all at once.
These tools allow you to:
Search for keywords, and contextual searches
Group similar documents
Spot recurring themes and data points
Large-scale data processors are great for large-scale investigations, as their name suggests. If you have data you could never process manually (or handle with an LLM), try these.
Example: OSINT with AI in Action
Imagine someone named Astra Velorin connects with you on LinkedIn. Their profile is mysterious: listing their title as ‘Ambassador for the Outer Spiral Arm’, and they’re offering you a job as an ‘Abduction Assistant’. Are they a genuine alien, or a particularly inventive scammer? Time to investigate - with AI.
Step One: Analyse the Profile
You paste Astra’s bio into an AI model - good old Chat GPT, in this case - and ask it to look for any other instances of the phrase “Ambassador for the Outer Spiral Arm” online. It brings up a sci-fi fan site, suggesting where the idea was lifted from.
Step Two: Generate Pivots
You write a clear, concise prompt: “show me a list of three follow-up checks to verify whether this profile is genuine.” ChatGPT returns a clear list of contextualised next steps:
Search for username reuse on gaming or role-play forums
Check if the “Interstellar Embassy” has a registered address
Reverse-image search their profile photos
You use image AI tools to follow-up on the latter - and find all the images come from an art subreddit.
Step Three: Examine Their Documents
To pique your interest in the position, Astra sends you a lengthy PDF titled First Contact Proposal. Instead of wasting time on reading it, you run AI analysis on the text, and find it’s an exact match for a well-known sci-fi novel. It’s also stylistically inconsistent with Astra’s LinkedIn posts, which are mostly about marketing.
Step Five: Summarise
You use your AI tools to create a clear summary of the investigation so far. Luckily - even though the investigation took you across the internet - everything is automatically archived. Leaving you with verifiable proof that Astra is not an extraterrestrial diplomat - just a bored nerd.
You reject the offer and remain on Earth.
Key Takeaways
Now you’ve met your new assistant! This issue should have taught you:
Your job is safe: AI won’t replace your investigative skills - just support them.
AI is dumb: It works best on straightforward tasks, with clear instructions.
Use many tools: each one is good for a different job.
Don’t waste your life: with AI, there’s no excuse to lose time on grunt work.
See you next issue, investigators!
✅ That’s it for the free version of The OSINT Newsletter. Consider upgrading to a paid subscription to support this publication and independent research.
By upgrading to paid, you’ll get access to the following:
👀 All paid posts in the archive. Go back and see what you’ve missed!
🚀 If you don’t have a paid subscription already, don’t worry. There’s a 7-day free trial. If you like what you’re reading, upgrade your subscription. If you can’t, I totally understand. Be on the lookout for promotions throughout the year.
🚨 The OSINT Newsletter offers a free premium subscription to all members of law enforcement. To upgrade your subscription, please reach out to LEA@osint.news from your official law enforcement email address.




This breakdown of proper AI prompting is incredibly valuable for investigators. The role-task-rules-output format you laid out is way more systematic than how I've been approaching it. I started using this structure yesterday for analyzing leaked datasets and saw imediately better results compared to my usual vague requests. The reminder that AI needs to be told not to guess is crucial - I've definitely been burned by hallucinations before when I wasn't explict enough.