What every law enforcement agency needs to know
Highlights
- Officers are already using AI faster than most agencies have policies to guide it.
- Consumer AI tools skip citations and audit trails, creating real risk for evidence and cases.
- A structured framework — covering data sourcing, courtroom defensibility, ROI, and documentation — helps agencies choose AI tools that hold up under scrutiny.
Artificial intelligence isn’t a future consideration for law enforcement anymore. It’s already inside the tools officers use every day to draw insights from public records. A recent Police1 webcast in partnership with Thomson Reuters “How to Evaluate AI Technology for Investigations” brought together Edgar Gallardo and Michael Murphy, both former officers now working for Thomson Reuters, along with Andrew Pellington, Senior Director of Product Marketing for Thomson Reuters, to unpack one practical question. How should agencies vet an AI tool before it touches an investigation?
Poll results from the session tell part of the story. 72% of attendees said their agency is already using or evaluating AI tools, and 87% said they believe AI is helping public safety. While AI has the potential to be one of the most valuable additions to agency workflows, not all tools are created equal. Agencies must carefully evaluate which solutions are secure, trustworthy, and supported by transparent citations.
To help navigate that process, the panel outlined a structured framework agencies can use to vet any AI vendor or solution provider before adoption.
Jump to ↓
| Where does the data actually come from |
| How consumer AI differs from purpose-built AI |
| Can the output hold up in court? |
| What return on investment can you expect? |
| Does it support your documentation and reporting needs? |
Where does the data actually come from
Every investigation starts with a source, and an AI-driven one shouldn’t be any different. Gallardo compared an unverified AI answer to an anonymous tip. You wouldn’t make an arrest on a tip alone, so why act on an AI output without knowing where the information originated? With many consumer tools, that origin is often an online forum or an unnamed dataset with no way to trace it back.
Murphy raised a related problem he described as “garbage in, garbage out.” Ask a general-purpose AI tool the same question weeks apart and the answer can shift, the same way an informant’s story can change under a second interview. If the underlying data or logic isn’t stable and traceable, neither is the conclusion an officer builds a case around.
What to ask a vendor
Before locking in a vendor, use these questions to pressure-test how the AI actually handles data. The answers will tell you whether you’re getting a tool built for evidentiary standards, or one that just sounds confident.
- Ask where each data point originates and how recently it was verified.
- Ask for a citation trail that links every output back to a named source.
- Confirm the same query returns consistent, sourced results over time.
How consumer AI differs from purpose-built AI
Consumer-grade tools are built for general use, without law enforcement data controls, citations, or audit trails. Murphy pointed out that these platforms often can’t show where an answer came from, and any personal data entered into them may sit on servers the agency doesn’t control.
Purpose-built investigative platforms take a different approach. They pull from curated, verified sources, cite the data behind every output, and date-stamp that information so investigators know how current it is. That distinction matters most under cross-examination, when a defense attorney only needs one weak link to challenge the whole case.
Can the output hold up in court?
An AI hallucination and a properly sourced AI output can look similar on the surface, but only one holds up once a prosecutor or defense attorney starts asking questions. An unverifiable output has no supporting documentation and can’t go in front of a judge, while a sourced output points directly to the public record, dataset, and date behind it.
Gallardo tied this back to the full lifecycle of a case. Search warrants, arrest documentation, and threat assessments are all facing more scrutiny than they used to, and defense strategies increasingly target the tools behind an investigation rather than just the conclusions. An audit trail that shows the query, the data pulled in, and the reasoning behind an output gives investigators something they can explain on the stand, not just a result they have to defend.
What return on investment can you expect?
Attendees reported spending five to ten hours a week researching suspects, witnesses, or persons of interest, a workload that Murphy said tracks closely with what he hears from agencies day to day. Gallardo framed the payoff in terms every investigator recognizes. A few recovered hours per week can go toward reviewing old reports or working a backlog instead of manual research.
Thomson Reuters CLEAR Investigate customers have reported research time reductions of up to 47%, turning hours of manual lookups into minutes.1 For smaller agencies working within limited budgets, the panel suggested asking vendors for real-world usage data that demonstrates measurable time savings and helps agencies assess whether the investment delivers meaningful value for their investigative workflows.
1. CLEAR Investigate Internal Study, Thomson Reuters, 2025.
Does it support your documentation and reporting needs?
Every step of an investigation needs to be written up somewhere, whether that’s a briefing note, an incident report, or a summary for a chief or watch commander. This kind of writing eats hours that could go toward active casework, but AI can turn hours of typing into a synthesized, professional summary in minutes.
Gallardo walked through how this plays out on a real case. A single phone number from a robbery victim can branch into associates, addresses, and prior reports, all of which need to connect back to their original source. Case closure matters, but so does being able to show — step by step — exactly how the investigation got there.
CLEAR Investigate in practice
Murphy shared how agencies are already combining CLEAR Investigate with two additional tools. Social Media Insights searches major platforms by phone number, email, or name, and Graylark Raven, a partner of Thomson Reuters, can geolocate an image down to a specific building or identify a vehicle from a partial photo. Used together, an investigator can move from a single photo or plate to a verified location or registered owner in a fraction of the time a manual search would take.
Bringing AI into an investigation isn’t a decision to make casually, but it also isn’t one agencies can put off. As Gallardo put it, officers already have AI in their hands whether or not a formal policy exists. Working through a structured framework, starting with data, moving through defensibility and ROI, and ending with documentation, gives agencies a way to bring AI in deliberately instead of reactively.
The right AI tool doesn’t just save time. It helps investigators find trusted information that stands up to scrutiny.
Discover how CLEAR Investigate helps agencies uncover connections faster with confidence.
CLEAR Investigate: Agentic AI for law enforcement
Accelerate law enforcement investigative research with AI designed for you
Unlock AI Potential ↗
Disclaimer
Thomson Reuters is not a consumer reporting agency, and none of its services or the data contained therein constitute a ‘consumer report’ as such term is defined in the Federal Fair Credit Reporting Act (FCRA), 15 U.S.C. sec. 1681 et seq. The data provided to you may not be used as a factor in consumer debt collection decisioning, establishing a consumer’s eligibility for credit, insurance, employment, government benefits, or housing, or for any other purpose authorized under the FCRA. By accessing one of our services, you agree not to use the service or data for any purpose authorized under the FCRA or in relation to taking an adverse action relating to a consumer application.
