When a client's AI tool breaks and they blame you
Here's how to handle the blame, fix the relationship, and protect yourself next time.
When a client's AI tool breaks and they blame you
Here's how to handle the blame, fix the relationship, and protect yourself next time.
It's a scenario more freelancers and agencies are running into every month: a client hires you to build or run a workflow using an AI tool, something goes sideways with the tool itself — a hallucination, a broken API, a sudden pricing change — and suddenly your phone is ringing with an angry client who thinks you broke it.
You didn't write the model. You don't own the servers. But right now, none of that matters to them. Here's how to handle it without losing the client, your reputation, or your mind.
1. Respond fast, even before you have answers
The worst thing you can do is go quiet while you investigate. Silence reads as guilt or incompetence. Send a short message within the hour — even if it just says: "I've seen the issue, I'm looking into it now, and I'll have an update for you by [specific time]."
That one message does a lot of heavy lifting. It shows you're on it, sets a clear expectation, and buys you time to actually figure out what happened. Clients who feel heard tend to calm down faster than clients who are left stewing.
2. Separate what broke from who's responsible
Once you've looked into it, write out a clear, plain-English summary of what actually happened. Was it:
- The AI tool's API going down?
- A model update that changed the output behavior?
- A third-party integration failing?
- A usage or rate limit the client hit on their own account?
Get specific. "The tool's underlying model was updated overnight and changed how it formats responses — that's what caused the error in your report" is a very different conversation than a vague "something went wrong." Specificity builds trust even when the news is bad. It also makes it obvious whether the failure was in your scope or the vendor's.
3. Know your contract before you get on the call
Before you jump on a call to hash things out, re-read what you actually agreed to deliver. Did you promise the tool would work, or that you'd build and maintain a workflow using the tool? Those are very different obligations.
Most solid service agreements include a clause that third-party services are outside your control and not covered by your SLA. If yours doesn't, now is a painful but important lesson: add that language before your next project. Going forward, be explicit in proposals — something like: "This workflow depends on [Tool Name]'s continued availability; outages or changes on their end are outside the scope of this engagement."
4. Fix what you can, document what you can't
Even if the failure is 100% the tool vendor's fault, look for something you can do. Can you build a temporary workaround? Switch to a backup tool? Manually handle the broken step while the vendor resolves it? Showing initiative — even small initiative — shifts the narrative from "my freelancer broke my project" to "my freelancer is fighting to keep my project running."
Document everything as you go: screenshots of error messages, vendor status pages, your own communication timestamps. If this ever escalates to a refund dispute or a bad review, that paper trail is your best friend.
5. Use this moment to reset expectations going forward
Once the dust settles, don't just move on. Have a brief debrief conversation with the client — even a five-minute voice note counts. Acknowledge what happened, explain what you're changing (if anything), and agree on how you'll handle future incidents.
This is also a good moment to introduce a simple incident protocol: you notify them within X hours, you provide a root-cause note, and you outline next steps. Clients who've been through one bumpy incident with a contractor who handled it well often become the most loyal long-term clients. The trust you build in a crisis is worth more than the trust built in smooth sailing.
Managing client communication when things go wrong is genuinely time-consuming — the updates, the documentation, the follow-ups. If you'd rather focus on fixing the actual problem instead of drafting careful messages at midnight, that's exactly the kind of work you can hand off to a Sidekyk. Just WhatsApp your AI sidekick the situation, and it can draft client-ready updates, pull together incident summaries, or even help you write tighter contract language for next time. Try it at sidekyk.ai.
Want this running in your WhatsApp every Monday morning?
Drop your number — we'll WhatsApp you the moment AI product teams goes live.
Join the waitlist on WhatsAppPowered by SideKyk · A team of AI agents in your WhatsApp