OSINT

It's the Second World War, and you need to know if the trains into Paris are still running. The problem is that Paris is behind enemy lines.

You could send a spy. You could send a reconnaissance plane through anti-aircraft fire and hope it comes back with photos.

Or you could just check the price of oranges.

Oranges reached Paris by train from Spain. When the railways were damaged, fewer oranges made it to the city's central market, and the price went up.

Try it below. Three rail lines run into the market. Click a line to bomb it, click it again to repair it, and watch the price.

Without ever seeing the tracks, the analysts could tell from the price how many lines were down.

And that's the whole idea behind open source intelligence, or OSINT. You take something anyone can see, ask it a good question, and learn something nobody ever announced.

In this essay, we'll look at what OSINT is, where it came from and how professionals do it. By the end, you'll know how to tell a solid signal from a lucky guess, and how to see your own online life the way a stranger might.

TL;DR

OSINT means using public information to answer specific questions.

There's no hacking and no secret sources. It comes down to patience, and to the fact that small public details add up to far more than anyone meant to reveal. That works on governments and companies, and it works on you.

What the name means

OSINT stands for Open Source Intelligence.

"Open source" here has nothing to do with software. It means information that's publicly or commercially available, such as a newspaper, a court filing, a satellite photo you can buy, a job posting, a restaurant review or a radio broadcast. If an ordinary person can legally get hold of it, it counts.

"Intelligence" is the word people skip over, and it's the more important one. The SANS Institute, one of the big training bodies in security, defines OSINT as intelligence produced by collecting and analysing public information for the purpose of answering a specific question.

Think of it like this.

Information is flour. Intelligence is bread.

A bag of flour won't feed anyone. A screenshot, a public record or a photo is just an ingredient until someone with a question mixes it, tests it and bakes it into something a person can act on. Collect a thousand screenshots with no question behind them, and all you've got is a messy kitchen.

There are a few other intelligence disciplines with similar names. HUMINT comes from human sources, SIGINT from intercepted signals, and GEOINT from imagery and maps. OSINT is the odd one out, because its raw material is already on your screen.

Older than the internet

You might assume OSINT started with social media. It started about two thousand years earlier.

In ancient Rome, the Acta Diurna was a daily public notice posted in the town square. Citizens read it. So could any visitor from an enemy city.

In 1862, during the American Civil War, Harper's Weekly printed details of a Union naval expedition, including the names of the ships. A Confederate officer could look up each ship, work out roughly how many troops and supplies the fleet was carrying, and guess where it might land. Generals on both sides read each other's newspapers as a matter of routine.

In World War I, British and French analysts read German newspapers and paid special attention to the obituaries. Counting dead officers was a way to measure losses Germany would never admit to.

The modern discipline even has a birthday. On February 26, 1941, President Roosevelt created the Foreign Broadcast Monitoring Service, an office that recorded, translated and analysed radio broadcasts from around the world.

Its first analytic report came out on December 6, 1941. It warned that Tokyo's tone was turning increasingly hostile.

The next morning, Japan attacked Pearl Harbor.

After the war, that small radio office was renamed and folded into the CIA. In 2005, on the recommendation of the 9/11 Commission, it was rebuilt as the Open Source Center. Today it's the CIA's Open Source Enterprise, where officers monitor foreign social and digital media and use AI to sort through it all. The discipline turned 85 this year.

Back in 1941, the raw material came through a radio speaker. Today, billions of people carry a camera, a GPS receiver and a publishing platform around in their pocket.

The mosaic effect

If you take one idea away from this essay, make it the mosaic effect.

A single tile from a mosaic is just a colour. Put enough of them together, though, and a face appears, even though nobody ever handed you the face.

Here's my favourite example.

In January 2018, an Australian student named Nathan Ruser was scrolling through Strava's Global Heatmap. Strava is a fitness app, and the heatmap showed every public run and ride its users had logged between 2015 and late 2017, which came to more than a billion activities.

Cities glowed. But over the deserts of Iraq and Syria, the map was almost completely dark, apart from a few bright little loops.

What were they?

Soldiers, jogging laps around the inside of their bases.

Each runner had only shared a workout. Put together, those workouts traced base perimeters, supply routes and daily routines. The Pentagon launched a review, and Strava tightened its privacy settings, all without anyone leaking or hacking a thing.

OPSEC, or operational security, is the practice of protecting small public details that could add up to something big. That makes it the mirror image of OSINT. An investigator is trying to assemble the mosaic, and a good security team is trying to make sure the tiles never line up.

Where the information comes from

From far more places than most people realise. Here's a rough list.

What people post on social media, forums, review sites and fitness or photo apps, which gives away locations, routines and relationships.

Public records such as court filings, company registries, property records and government contracts, which show who owns what and where the money goes.

Media coverage, from local news to press releases and trade magazines, which often carries details the national outlets miss.

Imagery from satellites, street-level cameras and user videos, which shows construction, troop movements, damage and environmental change.

Technical traces like domain registrations, website code, security certificates and devices left exposed online, which can reveal hidden infrastructure and links between websites.

Broadcast signals from ships and aircraft, picked up by public tracking networks and plotted on maps in close to real time.

And what analysts call behavioural exhaust, meaning prices, traffic, how busy a shop is and job listings. It's all the activity nobody meant to announce.

That last one is the most fascinating, and the easiest to misread. We'll come back to it.

How investigators actually work

The difference between a professional and someone taking screenshots at 2am is a process. One common version of the intelligence cycle has six steps.

  1. Ask. Start with a specific question, like whether a company's ship visited a sanctioned port last year. A sharp question tells you what to collect and, just as importantly, what to ignore. Trying to find out everything about a company tells you neither.
  2. Collect. Analysts separate passive collection, where you never touch the target, from active collection, where you interact with it. Passive is looking through a shop window. Active is walking in and asking the clerk questions, and now the clerk knows you were there.
  3. Process. Translate, timestamp, archive and organise. Save everything the moment you find it, because posts get deleted, accounts disappear and websites change overnight.
  4. Analyse. This is where the flour becomes bread. Cross-check every claim against independent sources, ask what else could explain what you're seeing, and look hardest for the evidence that would prove you wrong.
  5. Share. Give an answer in plain language, with an honest level of confidence attached.
  6. Repeat. A good answer usually raises a better question, which takes you back to step one.

Weather forecasters do this well. They'll give you a 70 percent chance of rain and leave you to decide whether to take an umbrella. Good OSINT answers work the same way, which is why they're full of words like likely, almost certainly and we can't confirm.

Buk 332

On July 17, 2014, Malaysia Airlines Flight MH17 was shot down over eastern Ukraine, killing all 298 people on board. Within hours, governments were blaming each other.

A small group of online investigators called Bellingcat started on something slower. They collected the photos and videos ordinary people had posted that day, including footage of a Buk anti-aircraft missile launcher moving through separatist-held territory.

Then came the patient part. They compared that launcher against photos posted online since 2010 by members of Russia's 53rd Anti-Aircraft Missile Brigade, based in Kursk. They matched seven distinguishing features, including paint marks and details on the side panels, and identified the vehicle as Buk 332. Later, they found a photo of the same launcher in a mechanic's personal album, taken before part of its unit number was painted over.

In September 2016, the Dutch-led international criminal investigation confirmed Bellingcat's findings about where the missile was fired from.

What cracked it was a few people comparing paint marks for a very, very long time. Small, boring details that anyone could check ended up carrying more weight than the confident statements coming from governments.

Bad signals

Washington has an old legend. The night before the first strikes of the 1991 Gulf War, pizza deliveries to the Pentagon reportedly jumped from a handful to over a hundred. Late nights mean takeout.

Today, the legend has a live dashboard. An account on X called Pentagon Pizza Report watches the "busier than usual" indicator Google Maps shows for pizza places near the Pentagon. On June 12, 2025, it flagged a surge at several of them around 7pm, hours before Israel launched strikes on Iran. The post went everywhere.

Then in August 2025, a Friday night spike fed baseless rumours online about the President's health.

So what happened?

Pizza places get busy for all sorts of reasons. A game on TV. Payday. The fact that it's Friday.

Every analyst has to learn to spot this trap.

If you remember the nights the pizza spiked before a war, and forget the hundred nights it spiked before nothing, you haven't built an indicator. You've built a horoscope.

Behavioural exhaust like this is great for noticing that something might be happening. Proving it takes other evidence.

Who uses OSINT

Pretty much anyone who wants to understand something someone else would rather keep quiet.

  • Journalists use it to verify whether a viral video was really filmed where and when the caption says.
  • Security teams use it to see their own company the way an attacker would. They check which employees list their job titles publicly, which old servers are still exposed and which passwords have leaked.
  • Attackers do the same research with the opposite goal. A convincing phishing email often starts with an org chart someone rebuilt from public profiles.
  • Businesses use it to check out partners, suppliers and acquisition targets before signing anything.
  • Activists use videos, satellite imagery and social posts to document war crimes, often for use in court.
  • Volunteers use it to help find missing people by gathering public leads on real cases and passing them on to law enforcement.

Every one of these groups is using the same skills, which is worth remembering when we get to the ethics.

The toolkit

Specific tools come and go every year, but the techniques underneath them stay the same.

The first is searching precisely. Search engines understand special instructions. Quotation marks force an exact phrase, and other operators can limit results to one website or one file type. Use them, and a search engine starts returning what you actually asked for instead of everything vaguely related.

Then there's reverse image search, where you search with a picture instead of words to find where else it's appeared. It's how you catch an old photo from another country being passed off as today's breaking news.

Geolocation means working out where a photo was taken from what's visible in it. Investigators study the shape of a mountain range, the style of a street sign, the language on a shop front, the type of power pole, and which side of the road the cars are driving on.

Chronolocation means working out when it was taken. Shadows work like a sundial. Their length and direction, combined with the location, can narrow down the time of day and even the season.

Metadata is the hidden information files carry around with them. Photos can store the camera model and sometimes GPS coordinates, and documents can store the author's name and when they were edited. The big social platforms usually strip it out, but plenty of other places don't.

Web archives are the internet's memory. Archive services regularly save snapshots of websites, so a page that's been edited or deleted might still exist in an older copy.

And signal tracking relies on the fact that large ships and most aircraft broadcast their position by radio for safety. Volunteer networks pick up those broadcasts and put them on public maps.

You don't need special access for any of this. You need curiosity and a lot of patience.

Where OSINT goes wrong

The same skills that identified a missile launcher can identify a stranger's home from a single selfie.

Stalkers, harassers and doxxers use these techniques. So do online mobs who think they're helping. After the 2013 Boston Marathon bombing, internet sleuths confidently named the wrong person as a suspect, and a grieving family paid the price.

Just because information is public doesn't mean every use of it is fair. Before acting on what they find, responsible investigators ask themselves a few questions.

Does the public interest actually justify looking into this person?

Am I collecting only what my question needs, or everything I can find?

Am I protecting the bystanders who show up in my material, by blurring faces and leaving out addresses?

And what happens to a real human being if I'm wrong?

If you can't answer those honestly, stop.

The verification problem

For most of OSINT's history, the hard part was finding information. Now, more and more, the hard part is trusting it.

AI can generate realistic photos, videos and voices quickly and cheaply. For a long time, the main test of a piece of footage was whether it looked real. Now that fakes can look flawless, investigators are asking different questions. Where did this file come from? Who posted it first? What else, from a completely separate source, agrees with it?

A single video counts for a lot less these days. It has to line up with satellite imagery, eyewitness accounts from different sources, weather records, and shadows that match the claimed time.

To be fair to AI, the same technology flooding the internet with fakes is also helping analysts sort through more material than any human team could ever read.

Source grading and other habits

If you already do this work, these are the habits that keep a solid finding from turning into a viral mistake.

Grade the source and the claim separately. The Admiralty system, used widely in intelligence and policing, rates a source's reliability from A to F and the credibility of the information from 1 to 6. A historically reliable source can still pass along a bad claim, and a random account can still post a real video. Keeping the two ratings apart stops a source's reputation from doing your thinking for you.

Test competing hypotheses. Before you settle on an answer, write down every plausible explanation and check each piece of evidence against all of them. The strongest explanation is often the one with the least evidence against it, rather than the most evidence for it.

Be suspicious of the moment you think you've found it. The first plausible match is the most dangerous one, because relief switches off your scepticism. Slow down exactly when you feel most certain.

Preserve everything as if it'll end up in court. Capture the original file, the page around it, the URL, the time of capture and a cryptographic hash. The Berkeley Protocol, published in 2020 by the UN Human Rights Office and UC Berkeley's Human Rights Center, sets out professional standards for digital open source investigations, and it's worth reading in full.

Watch your own footprint. Some platforms tell people who viewed their profile, and some websites log their visitors. Research carelessly and you can tip off the very person you're investigating.

Write down what you didn't find. A documented dead end is evidence too, and it saves the next person from walking into it.

Check yourself

You don't have to become an investigator for any of this to matter, because you're already a source. So tonight, try this.

  1. Search your full name in quotation marks. Then add your city or employer, and see what a stranger would find.
  2. Reverse image search your profile photo. It shows where else the photo appears and which of your accounts it ties together.
  3. Turn on route hiding in your fitness apps. Most of them can hide the start and end of your runs, so your front door isn't on a map.
  4. Read your public posts as if you were a stranger. Do they give away your gym schedule, your kids' school, or the fact that you're on holiday right now and your house is empty?
  5. Look up your email address on Have I Been Pwned. Change the password for any account that shows up in a breach.
  6. Find the old accounts you forgot about years ago. They're usually the least protected and the most revealing.
  7. Before you post a photo, check the background. House numbers, number plates, post on the table and the view from your window can say more than the photo itself.

Think of it as choosing which tiles of your mosaic you leave out on the table.

Back to the oranges

All the analysts watching Paris needed was a good question and the patience to look at something ordinary from a new angle.

OSINT still works that way. The world reports on itself all day, through prices, photos, shadows, workouts, takeout orders and ship signals, and most of it goes unnoticed until someone asks the right question.

So the next time you post something, remember that somebody, somewhere, might be checking the price of your oranges.