Customer service teams: Are customers using AI before contacting you? We're seeing a major shift in support queries

Jennifer, this is happening across the industry. We’ve been studying it.

The new customer journey:

Old: Problem → Google → Company Help Center → Contact Support New: Problem → ChatGPT → (Maybe) Company Help Center → Contact Support

What this changes:

1. Simple queries deflected - AI answers the easy stuff
2. Complex queries remain - AI can’t handle edge cases
3. Pre-researched customers - They’ve already tried things
4. Misinformed customers - AI gave wrong info

The data from our support:

Fewer tickets, but each one takes longer because the easy ones are gone.

Knowledge management perspective on the AI customer service shift:

Your help content is now training AI.

What’s in your help center, documentation, and FAQs is what AI learns about your product. If your content is:

• Incomplete → AI fills gaps with guesses
• Outdated → AI provides old information
• Unclear → AI misinterprets

The solution:

Treat your help content as AI training data. It needs to be:

1. Comprehensive (cover all features)
2. Current (update regularly)
3. Clear (unambiguous language)
4. Correct (factually accurate)
5. Explicit about limitations (what you DON’T do)

What we changed:

We added sections like:

• “What [Product] does NOT do”
• “Common misconceptions about [Product]”
• “Differences between [Product] and [Competitor]”

This helps AI provide accurate information BEFORE customers contact support.

Operations perspective on the shift:

Staffing implications:

If simple tickets decrease and complex tickets increase, you need:

• Fewer tier 1 agents
• More tier 2/3 specialists
• Different training (complex problem solving vs. process following)
• Longer handle time expectations

How we adapted:

1. Reduced tier 1 team by 20%
2. Promoted best performers to tier 2
3. Changed success metrics (handle time → resolution quality)
4. Created “AI escalation” workflow for misinformation cases

The cost reality:

Lower volume but higher complexity = roughly same total cost BUT customer satisfaction increased because fewer simple queries means less queue wait time for complex ones.

Content strategy to reduce AI misinformation:

The problem: AI is a black box - you can’t directly correct it. But you CAN influence what it learns.

What we do:

1. Comprehensive FAQ - Every common question answered clearly
2. Explicit limitations - What we DON’T do, clearly stated
3. Pricing structured data - Current pricing in schema markup
4. Feature descriptions - Clear, unambiguous language
5. Comparison content - How we differ from competitors

Monitoring:

We use Am I Cited to track what AI tells users about us. When we spot misinformation:

1. Create/update content addressing it
2. Add to FAQ if it’s a common question
3. Wait 4-8 weeks for AI to learn the correction
4. Monitor for improvement

It’s not instant, but you can systematically correct AI’s understanding of your product.

We actually built AI into our support workflow. Here’s the impact:

AI-assisted support model:

1. Customer starts chat
2. AI bot handles first contact
3. If AI can’t resolve, escalates to human
4. Human sees AI’s attempted solutions

Results:

The key insight:

By the time a customer reaches a human, they’ve already:

• Described their problem to AI
• Had AI attempt solutions
• Confirmed what doesn’t work

Human agents start with full context. More complex, but more efficient.

Customer research perspective:

We surveyed 500 customers about their AI usage before contacting support:

The “AI-first” customer segment:

They’re typically:

• Tech-comfortable
• Prefer self-service
• More frustrated when they DO contact support (because “simple” solutions failed)
• More specific in their problem descriptions

Implication:

When they reach you, they’re often further along in frustration but also better at describing the problem.

Training perspective on handling AI-influenced customers:

New skills our agents need:

1. AI-awareness - Understanding what AI can/can’t do
2. Misconception handling - Correcting without shaming
3. Context gathering - “What have you already tried?”
4. Documentation skills - Logging AI-related issues
5. Escalation judgment - Knowing when AI misinformation needs content update

Training modules we added:

• “Understanding the AI-first customer”
• “Handling AI misinformation gracefully”
• “What AI tells customers about our product” (based on Am I Cited monitoring)
• “Logging patterns for content improvement”

The cultural shift:

Agents now see themselves as part of a feedback loop. Their observations about AI misinformation flow to content team, which updates docs, which improves AI accuracy.

This thread validated what I suspected and gave me actionable strategies. Key takeaways:

The reality:

• AI is deflecting simple queries (15% fewer tickets)
• Complex queries are increasing (+50%)
• Handle time is increasing (simpler stuff is gone)
• Misinformation creates new challenges

Strategies to implement:

Short-term:

1. Create “AI misconception log” for agents
2. Train team on handling AI-influenced customers
3. Adjust success metrics away from pure handle time
4. Start monitoring what AI says about us

Medium-term:

1. Update help content to be “AI training friendly”
2. Add explicit content about what we DON’T do
3. Create feedback loop from support to content team
4. Consider AI-assisted support model

Long-term:

1. Restructure team for complex query handling
2. Shift hiring toward problem-solving skills
3. Build systematic AI information monitoring

The survey data showing 62% use AI first is significant. This isn’t a trend - it’s the new normal.

Thanks everyone for the operational and strategic insights.