Investigator AI that shows its evidence
An AI tool belongs in an investigation only if you can check what it tells you. Oceanir is an AI for one job within investigative work: working out where a photo or video frame was taken. It shows the places it weighed, the clues for and against each one, and what it could not settle, so the checking stays with you and the conclusion stays yours.
Try a free location check. Usage limits apply; sign-in may be required. Your photo stays attached.
Written by the Oceanir team. Last reviewed 2026-10-26.
Where AI helps an investigation, and where it should stop
AI is good at the tedious, high-volume part of image work: reading signs and scripts in several languages, noticing that a road marking or a pole design points to one country rather than another, and comparing a scene against many possible places faster than a person can. Those are the steps where it saves hours.
It should stop at the decision. An AI can propose where an image was taken. It cannot know that the file you hold is the original, that the scene has not changed since the photo was taken, or that the picture was not staged. Those judgments need a person who is accountable for the finding, and a good tool is built to hand the decision back.
Seven questions to ask any AI tool before it touches a case
Vendors describe their tools in terms of speed and accuracy. For case work the better test is whether the tool lets you inspect its reasoning. Put these questions to any AI product, including this one.
01Does it show the clues it used?
A bare answer is a guess you cannot check. You want the specific things in the image that led to it, such as a sign, a road marking or a building outline, so you can look at them yourself.
02Does it argue against its own answer?
Look for the reasons a candidate might be wrong. A tool that lists only support for its top answer is leaving the hard part to you without saying so.
03Does it say what it could not settle?
A result that admits it only supports a region is more useful than a confident pin that is wrong. Check that the tool states how precise its answer is.
04Is its confidence figure honest?
A confidence score describes the tool's own reading of the evidence. It is not proof about the world. Be wary of any score that is presented as verification.
05Can you repeat and record the result?
You need to show later what you asked, what came back and when. Saved results and exports matter more than they sound once a case is reviewed.
06Does it have limits you can read?
A tool that will analyze anything for any purpose is a risk to a case file, not a feature. Look for a published statement of what it will not do.
07Who makes the final call?
The finding should be yours. The tool should support a decision, and your write-up should say which parts a person confirmed by hand.
What an Oceanir result contains
A deeper check returns ranked candidate places, not a single pin. For each one you see the clues that support it and what argues against it, so you can see why the second-ranked place lost to the first. The result states how precise it is, whether an area, a district or a specific site, and a photo that supports only a region is reported as a region.
The confidence shown is Oceanir's own reading of the scene, and the result says it does not verify a location. Results are saved to your history, you can add them to a case, and Pro includes PDF and CSV exports for your notes. Teams that handle many images can run the same checks through the developer API or the MCP tools.
A routine that keeps the person in charge
AI-assisted work is easy to defend when each step is separate and written down. This is a routine that fits any tool, and it is the one Oceanir is designed around.
01Keep the original and note its origin
Do this before you run anything, so the analysis always points at a known file.
02Run the check, then read the case against it
Open the top candidate and read what argues against it before you accept it. If the objections are strong, move to the next candidate.
03Confirm on independent imagery
Compare the best candidate with street-level or satellite views you found yourself. Match fixed features such as rooflines and kerbs, not shop fronts.
04Write down what a person verified
Separate what the AI proposed from what you confirmed. A reviewer should be able to see where the tool stopped and you began.
05State the limit of the finding
Say whether the image supports a street, a district or only a region, and what you could not rule out.
What this does not do
Oceanir works on places and scenes. It reads what is visible in an image to estimate where it was taken. It does not read the GPS tags saved in a file, and it does not tell you whether an image is authentic. An exact location is not guaranteed, and some images do not contain enough to place at all. Its boundaries are published on the product boundaries page.
Frequently asked questions
What is investigator AI?
It is software that uses artificial intelligence to speed up part of an investigator's work, such as reading images, sorting evidence or drafting notes. The useful versions show how they reached a result and leave the decision with the investigator. Oceanir does this for one task, working out where a photo or video frame was taken.
Can AI replace an investigator?
No. It can narrow where to look and surface details a tired reader might miss, but it cannot establish that a file is the original, that a scene is unchanged or that an image was not staged. A person has to verify the result and stand behind the finding.
Is an AI result evidence I can rely on in a report?
Treat it as a lead to verify, not a certified finding. Confirm the place against independent imagery such as street-level or satellite views, and write down what you checked and what you did not.
How should I document AI-assisted work?
Record the tool and the date you ran it, a hash of the file you analyzed, the result you received, the imagery you used to confirm it, and the points a person checked by hand. Say plainly where the AI was a starting point and where your own checking began.
Can I use it from a script or an agent?
Yes. Oceanir has a developer API and MCP tools for teams that process many images or want the check inside their own workflow. The same rules apply: the output is a lead with its clues attached, and a person reviews it.
What happens when an image has too little to go on?
The result says so. It reports a wider area, or that the image could not be placed, instead of inventing a pin. That honest answer is useful in a case, because it tells you to look for more material.
Related
Run one image and read the case against it
Start with a free scan, then open a deeper check and read what argues against the top answer before you accept it.
Start free