The AI Content Bottleneck No One Warned You About

Last Tuesday, I had twelve browser tabs open, all Google Docs, all AI-generated blog posts in some state of review. We used my AI writing agent to draft a wave of content in about four hours.

I was supposed to feel liberated. Instead, I feel buried.

If you’re running marketing for a technology company, this probably sounds familiar.

We Fixed the Wrong Problem

For years, the conversation was about capacity. We needed more writers, more designers, more hands to make the stuff. AI was supposed to solve this. And it did. My team can now generate more content before lunch than we used to produce in a week.

But here’s what nobody is talking about: all that content still needs someone to check if it actually makes sense. Accurate, on-brand, strategically sound, and not accidentally promising features our product doesn’t have or linking to facts from a competitor’s survey.

The bottleneck didn’t disappear. It just shifted.

The Part Where AI Confidently Lies to You

AI writes with the confidence of a consultant who definitely didn’t read the brief.

Two weeks ago, our AI drafted a blog post about the reliability of an open source project. Well-written, good flow, right terminology. It also described in painstaking detail all the weaknesses of this technology that underpins our product. While this AI agent dramatically accelerated the outline and first draft process, it missed the fact that a deft hand is needed to subtly explaining a pain point vs. bashing a technology.

A writer might put “[NEED TO VERIFY]” next to something they’re unsure about. AI just keeps going. Remember AI is probabilistic at it core and will often share things that are highly likely and not always certain. It’ll invent statistics, misrepresent your positioning, and embellish case studies.

Because it sounds authoritative, it’s easy to miss. You can’t skim. You have to actually read it closely.

Multiply that by twelve browser tabs, and you see my problem.

Why This Gets Worse as You Scale

Last month:
- Time generating content with AI: ~4 hours
- Time I spent reviewing: ~16 hours
- Content that didn’t get stuck with a reviewer so we actually published: 30% of what was drafted

It is not unreasonable to plan for a 4:1 review-to-creation ratio.

We’re struggling to realize the promise of generative AI, especially at scale. I used to think that it was about the technology. Now I think everyone’s drowning in drafts they can’t review fast enough or get stuck because some “expert” doesn’t like it.

What I’m Trying Instead

The big shift: I stopped treating all AI content the same way.

Some stuff just doesn’t matter that much. Weekly newsletters, social posts, internal updates—these don’t need 45 minutes of fact-checking. They need a quick brand voice check and a “does this sound obviously wrong?” scan.

Does it sound like us? Any sketchy claims? Are the links right?

I built these checks into my writing agent to find the most egregious items.

Other stuff absolutely matters. Thought leadership, product positioning, anything customers use to make buying decisions, technical content where accuracy isn’t optional.

These still come to me, but we’re validating against specific questions:

• Is this technically accurate?

• Does this support our positioning?

• Would this be embarrassing if a competitor screenshotted it

• Is there a genuine insight here, or just remixed content?

If it fails these tests, I don’t fix it. I send it back.

This felt wasteful at first—rejecting something 70% done. But rewriting a mediocre AI draft takes longer than generating a new one with better direction.

The Rejection Rate Nobody Talks About

We’re rejecting about 30% of AI-generated drafts outright. I just reviewed a draft today where AI smashed us in the first paragraph. I almost fell off my chair.

This is actually the sign of a healthy process. If you’re approving everything AI generates, you’re probably publishing mediocre content (slop).

You need to get comfortable with “waste.” AI can generate ten versions in the time it took to write one. That’s only valuable if you’ll throw out the seven that aren’t good enough.

What Actually Changed

I don’t think AI made my job easier. It made it different. Less time writing, more time judging. Less execution, more curation.

The teams figuring this out aren’t generating the most content. They’re the ones who redesigned their workflow, who approves what, how review works, what quality bars apply to different content types.

We’re not as fast as the “AI will 10x your content output!” headlines promised. But we’re publishing better stuff that meaningfully supports the customer journey.

If You’re Dealing with This Too

Track it. Measure how long AI takes to create vs. how long you spend reviewing. If you are at V9 of something, then maybe you have a problem…

Sort content into “high stakes” and “low stakes” buckets. Be honest about what needs your personal review.

Build a checklist for each bucket. Not an editing guide but a validation framework.

Get comfortable rejecting stuff. If it takes 30 minutes to fix an AI draft, have AI try again.

The bottleneck will always exist somewhere. The question is whether it’s the “Review” column of your Kanban board or something else you can manage.

Right now I’ve got 7 tabs open. Down from twelve.

Progress.

---

*How are you handling AI content review? I’m still figuring this out and would love to hear what you are experiencing

---