How AI changed my approach to QA

There’s a version of me that will always argue that content comes first. But when you’re reviewing ten writing tests in a single day, alongside your actual job, speed stops being a nice-to-have and becomes the only way anything gets done at all. However, I didn’t start integrating AI into my QA process till well after I had to speed up something completely unrelated. And, for transparency, I would like to make it clear that leveraging AI in any form in content has been one of my biggest pet peeves for the past 3 years. I know I am not alone in that. What’s ironic today, though, is how many LinkedIn posts from all avenues are extremist in their opposition to AI in content. That is a post for another day, perhaps; I have big opinions on this. But, I digress.

The moment I started integrating AI in my own processes was during a hiring crunch. Scaling a team at the speed of light poses many challenges. Some days, I had well over ten tests to review between daily tasks, including editing, amid the same hiring crunch that I’ve written about elsewhere. What helped the overall QA rounds was the checklists I had in place, which let me know exactly what to look for. With workload increasing and the need to scale the team, I had to come up with a solution, fast. So, I caved. The one person who was against AI started experimenting with how AI could speed things up, without taking away the human eye (the devil is in the detail, after all). 

So, I drafted a prompt that worked on the same checklist I had created, and ran it through AI while I did the human bits, proofreading. Instead of just letting AI ‘do the work’, both myself and AI worked on the same piece of content at the same time, looking at it from different angles. The necessity of human oversight in content is unquestionably one hill I am prepared to die on (oh how I hate that expression). 

This new process helped me work through tests that would usually take me hours way faster, and also why I started running this process in everyday QA instead of using AI only for the humdrum.

What QA looked like before AI

My ‘traditional’ QA process was a fully manual pass, which took around one hour per content piece depending on quality:

• Quick proofread to get a feel for the piece as a whole, marking obvious issues

• Return to flagged issues, manually correcting them

• Compare the brief and SEO instructions with the delivered work and flag next issues

Issues I focused on in a nutshell:

• Spelling

• Grammar

• Factual accuracy

• Keyword integration

• Link integration

• Logic

• Structure

• Headings

If the cross-checking and fixing process took longer than the designated hour and the overall quality warranted revisions, the piece was sent back to the writer with actionable feedback to ensure recurring errors were nipped in the bud.

This process worked well for me, and resulted in high-quality output and writers who were coached to do their best. It did, however, cause bottlenecks.

What actually changed

I'd already been using Gemini for general grunt work before it ever touched QA, the kind of small, repetitive tasks that eat time without needing real judgment. Once I started running it against my actual QA checklist, the shift was real: I'd do my own proofread as before, and in parallel, AI would check the same piece against the important, well-defined criteria and hand back a list of the major issues. Two passes happening at once instead of one long pass alone. Turnaround time was cut in half.

On the writing side, I'd also had writers using NeuronWriter to optimize their own drafts before submission. That didn't remove my job; it just moved it: I'd go in afterward and check what they'd missed, which, more often than not, was something. NeuronWriter caught the surface-level gaps. It didn't catch everything, and it never fully replaced a second set of eyes.

Over time, the AI side of the process actually got sharper. It got better at pinpointing inconsistencies within a piece, not just flagging keywords, but catching logic that wasn’t pulling the piece together coherently.

Where AI still lets me down

Factual accuracy is the one area AI never earned my full trust. It's still, to this day, something I check manually every time, no exceptions. I don't have one dramatic example of it getting something wrong; it's more that the risk of it being subtly, confidently wrong is high enough that skipping the human check was never something I was willing to do. Grammar, structure, keyword logic, AI handles that well. Whether a claim in the content is actually true still requires someone who knows the subject.

Teaching a team to use AI, discerningly

Introducing AI into QA wasn't just a change for me; I had to actively train writers and guide team leads through it too. The clearest example: a lead reached out, stuck, not sure how to shorten or restructure a blog post that had gotten far too long, while still meeting the SEO instructions we'd been given for that piece. I wrote a prompt that gave AI the exact outcome I wanted, specific on length, specific on structure, and had a cleaner, tighter version of the post in less than ten minutes.

I didn't just hand the fix back. I shared exactly what I'd done, the actual prompt and my reasoning behind it, so the lead had something repeatable next time they hit the same wall. That's really the difference between using AI well and just using AI: it's not about the output of a single prompt; it's about building a skill the team can reuse.

How I'd build a QA process today

If I were setting up QA for a new team from scratch now, I wouldn't design it as "human QA, then AI QA" or the other way around. I'd build it the way it actually ended up working: parallel, a checklist-driven AI pass running alongside a human proofread, each doing what it's genuinely good at. AI for consistency, structure, and the defined mechanical checks. A person, always, for whether the content is actually true.

The tool changed the pace. It didn't change what the job is actually for.