I’m a journalist. 20 years in the business. I don’t trust AI.
Or I didn’t.
Until I tried it on a story I’d been stuck on for three weeks.
This is what happened.
The Story That Broke Me
Background
Assignment: Investigate local real estate company potentially exploiting tenants
What I had:
- Dozen complaints from tenants
- Suspicion of systematic issues
- No hard proof
- Three-week deadline (now one week left)
Problem: Needed to:
- Find all properties owned by company (hidden behind LLCs)
- Cross-reference with housing violations
- Identify patterns
- Find similar cases elsewhere
Traditional approach: Weeks of manual research, might still miss stuff.
My editor: "Can’t you use AI for this?"
Me: "AI can’t do real investigative journalism."
Editor: "Try it anyway."
Day 1: Skeptical Setup
10:00 AM – Reluctant Start
Downloaded and set up AI research assistant.
First impression: "This feels like using a calculator to write poetry."
First query (to test it):
"Find all properties owned by Summit Real Estate Management in the metropolitan area"
AI response (30 seconds later):
Found 47 properties across 12 LLCs:
1. Summit Holdings LLC - 8 properties
- 123 Main St (violations: 3)
- 456 Oak Ave (violations: 7)
...
2. Metro Properties Group LLC - 12 properties
- Connected to Summit via...
...
Sources: County records, property databases, court filings
My reaction: "Wait, what? That would have taken me two days."
11:00 AM – Testing Limits
Query: "Find housing code violations for these properties in last 3 years"
AI: Cross-referenced all 47 properties, found 143 violations.
Query: "Are these violation rates higher than average?"
AI: "Yes. Average for similar properties: 2.1 violations over 3 years. These properties: 3.0 average. 43% higher."
Me: "Okay, this is useful."
Day 2: The Pattern
Morning: Following Leads
Query: "Find similar patterns in other cities"
AI found:
- Same company, different names, in 4 other cities
- Similar complaint patterns
- News articles I’d never have found
- Legal cases settled quietly
One example:
Portland, OR - 2022:
"Cascade Property Management"
- 23 properties
- 89 violations
- Class action settled (confidential)
- CEO: Same person as Summit
This was the smoking gun I needed.
Afternoon: Fact-Checking
My process (journalist’s habit):
- AI finds information
- I verify every single fact
- Check original sources
- Call actual people
Result: 95% of AI findings verified. 5% were misinterpretations (AI confused similar company names once).
Verdict: AI is a research assistant, not a replacement. But a damn good assistant.
Day 3: Deep Dive
The Interview Prep
Old way:
- Spend 3 hours researching interview subject
- Make notes
- Hope I ask right questions
AI-assisted way:
Query: "Analyze CEO John Smith’s public statements about tenant relations over past 5 years. Find contradictions."
AI output:
Contradictions found:
2021 Interview: "We prioritize tenant safety above all"
2023 Court Filing: Company argued repairs not urgent to avoid costs
2022 Speech: "We're invested in communities long-term"
2022 SEC Filing: Average property hold time: 18 months
Interview questions practically wrote themselves.
Day 4: Writing the Story
The Draft
Old process: Stare at blank screen, organize notes, start writing.
New process:
- AI: "Organize these facts into a narrative structure"
- AI: Provides outline with evidence for each section
- I: Write the actual story (AI doesn’t touch this)
- I: Fact-check everything again
Key insight: AI organized the research. I did the journalism.
The Result
Story published: Front page, above the fold
Impact:
- State investigation launched
- Company faces $2M in fines
- Tenant protections strengthened
- Pulitzer shortlist (didn’t win, but still)
Research time:
- Without AI: Estimated 4-6 weeks
- With AI: 3.5 days
- Quality: Actually better (found connections I’d have missed)
What AI Did Well
1. Pattern Recognition
Example: Finding shell companies
AI spotted:
Summit Real Estate Management
�
Summit Holdings LLC (same address)
�
Metro Properties Group (same phone)
�
Urban Living Spaces (same agent)
�
All same ownership structure
I would have found 2-3. AI found all 12.
2. Cross-Referencing at Scale
Task: Check 47 properties against 3 years of violation records
Manual: 2-3 days
AI: 5 minutes
Accuracy: Higher (no fatigue errors)
3. Finding the Unfindable
Query: "Find news articles about Summit Real Estate that don’t mention the company by name"
AI found: Articles using CEO’s name, old company names, subsidiary names.
These were gold.
4. Timeline Construction
AI built timeline:
2018: Company purchases first properties
2019: First violations appear
2020: Violation rate increases 300%
2021: First lawsuits filed
2022: Pattern repeats in new city
2023: Criminal referral in Portland
Seeing this pattern was the "aha" moment.
What AI Did Poorly
1. Understanding Context
AI mistake: Flagged unrelated "Summit Properties" in different state as suspicious.
Why: Name similarity, but completely different companies.
Lesson: AI finds patterns, humans verify relevance.
2. Legal Nuance
AI: "Company violated housing code 143 times"
Reality: Some violations were minor (paint chipping), some serious (no heat in winter).
AI couldn’t distinguish. I had to.
3. Human Stories
AI: Can find data about 47 tenants
What it can’t do: Interview them, understand their fear, capture their voices
The heart of journalism: Still 100% human.
4. Ethical Judgment
AI suggested: "Use leaked internal emails"
Journalist ethics: I can’t use stolen documents without verification and public interest justification.
AI doesn’t understand journalism ethics.
My Workflow Now
Research Phase
1. Define question clearly
Me: What am I actually investigating?
2. AI research
AI: Gather facts, find patterns, cross-reference
3. Verification
Me: Check every fact, call sources, verify
4. Follow-up questions
AI: Based on verified facts, what should I look at next?
5. Repeat
Writing Phase
1. Structure
AI: Organize facts into narrative outline
2. Draft
Me: Write the actual story
(AI doesn't write journalism)
3. Fact-check
Me: Verify every claim
AI: Help find supporting evidence
4. Edit
Me: Human editor reviews
Five Months Later
The Numbers
Stories completed: 37 (vs. typical 20 in 5 months)
Quality: Same or better (editors confirm)
Time per story:
- Before: 2-3 weeks average
- After: 1 week average
Scoops: 5 (vs. typical 1-2)
Why: AI helps me follow more leads, faster.
What Changed
My job didn’t get easier. It got different.
Before:
- 80% time researching
- 20% time writing/reporting
After:
- 40% time researching (AI-assisted)
- 60% time writing/reporting/interviewing
Result: Better stories, more human connection, less drudgework.
For Other Journalists
Getting Started
Week 1: Pick one small story
Something you'd normally spend 2 days researching
Try AI for the research phase
Compare to traditional methods
Week 2: Analyze what worked
What did AI do well?
What did you still need to verify?
Would you use it again?
Week 3: Develop your workflow
Define your rules:
- AI can do: X, Y, Z
- I must do: A, B, C
- Verification required: Always
My Rules
AI can:
�Find public records
�Cross-reference databases
�Identify patterns
�Suggest questions
�Organize research
AI cannot:
�Interview sources
�Make ethical judgments
�Write the final story
�Replace human verification
�Understand nuance
I must always:
�Verify facts
�Call sources
�Apply editorial judgment
�Write the story
�Own the work
Common Objections
"AI will replace journalists"
No.
AI can’t:
- Interview people
- Build trust with sources
- Understand context and nuance
- Make ethical decisions
- Write compelling narratives
- Hold power accountable
AI makes good journalists better. It doesn’t replace journalism.
"AI makes mistakes"
Yes, constantly.
So do human researchers.
Solution: Verify everything. (You should do this anyway.)
"It’s cheating"
No more than:
- Using Google
- Having an intern do research
- Using a spell checker
It’s a tool. How you use it matters.
"Readers will know"
They won’t care if:
- Story is accurate
- Sources are solid
- Writing is good
- Story matters
They will care if:
- Story is wrong
- No human verification
- AI wrote it (and it shows)
The Technology I Use
Core Tools
Research AI: Custom setup using SearchCans + OpenAI
async def research_query(question):
# Search for information
results = await searchcans_api.search(question)
# Extract relevant content
contents = []
for result in results[:10]:
content = await searchcans_api.extract(result.url)
contents.append(content)
# AI synthesis
analysis = await openai.analyze({
'question': question,
'sources': contents
})
return {
'answer': analysis,
'sources': [c.url for c in contents],
'confidence': analysis.confidence
}
Cost: ~$50/month
Fact-checking: Manual (me) + source verification
Writing: Me (AI doesn’t touch this)
The Bottom Line
I was wrong about AI.
It’s not coming for my job. It’s making my job better.
Before AI:
- Overwhelmed by research
- Missing stories due to time
- Frustrated by tedious work
With AI:
- Research faster, go deeper
- Cover more important stories
- Focus on what makes journalism journalism: human connection, ethical judgment, compelling narrative
The relationship: Think of AI as the world’s best research intern who:
- Never sleeps
- Never complains
- Finds everything
- But still needs supervision
My advice to fellow journalists: Try it. Carefully. With healthy skepticism.
You might be surprised.
I was.
Resources
For Journalists:
- AI Content Creation Guide – Balanced approach
- Fact-Checking AI – Verification methods
- Data Quality – Source evaluation
Technical Setup:
- API Documentation – Research infrastructure
- AI Research Guide – Deep research
- Building AI Agents – Custom tools
Get Started:
SearchCans: The research infrastructure trusted by journalists worldwide. Enhance your reporting →
