TL;DR: AI is not going to replace writers, content strategists, or anyone else. General-purpose AI can’t perform content audits. It can, however, improve them. In this post I rant a bit, rave a bit, and then talk about how I use AI to improve (not replace) my content auditing toolset.

If you don’t want to hear me rage about marketing’s never-ending, swirling plunge down the trendy technology whirlpool and want to stick to the how-to, skip this next bit.

I rant

Stop it.

This is my “Stop it” face

This is my second post about AI that says “Stop it.” Get the hint, people.

If you’re talking about AI coming for our jobs, stop it. If you’re calling for “greater efficiency” when you mean “fire the content team and replace them with bots,” stop it. If you’re rolling your eyes in a here-goes-Ian-ranting-again-he’s-old-ignore-him way, also stop it. I know whereof I speak. I’ve seen some stuff. General-purpose AI is going to replace:

1. Jack
2. S—, uh, poop

Yes, AI is revolutionary. We’ve never seen anything like it. It’s fun. It’s also destroying the planet, but hey, who isn’t?

In marketing, general-purpose AI can’t replace anything. Let me adjust that a bit: General-purpose AI can’t replace anything.

There are lots of reasons for this:

1. John Cass points out that, if we use AI to do content audits at scale, cost is a constraint. If you use, say, GPT to extract content, parse it, then analyze it, page-by-page, yeah, it’s going to cost you a fortune
2. General-purpose AI is going to do an awful job. It can’t audit content, because it doesn’t understand words. If you want to dive deep into this question, read Veena D. Dwiviedi’s piece on the topic
3. No one with a shred of business sense or marketing expertise is going to trust a massive model based on English-language writings by a bunch of cellar-dwelling nerds, random white dudes, internet trolls, college professors, and 20th-century journalists to create compelling marketing. Professors, journalists, nothing but love. But you gotta admit it’s a stretch
4. General-purpose AI is just that. General-purpose. It has no special marketing knowledge. Just the accumulated ravings of every attention-starved marketer trying to grab the next conference keynote

How I use AI to improve content audits

We can be more efficient, though. Here’s how I add AI into the mix:

1. You: Write your personas and map the user journey. Sorry, AI can’t do that (yet)
2. ScreamingFrog: Crawls your site. I like ScreamingFrog because it’s a nice, portable desktop crawler that can still crunch its way through huge sites. You can also use Sitebulb or another cloud-based crawler
3. ScreamingFrog: Extracts essential content like H1s and paragraph text, while it removes stuff like navigation
4. ScreamingFrog: Calculates the Flesch Reading-Ease score
5. Google Analytics API: Gets page-by-page performance data. Hey! Guess what! Most crawlers – like ScreamingFrog – connect directly to the Google Analytics API and will grab this data for you Not using Google Analytics? Not a developer? Export it all and use VLOOKUP. Last year, I talked about that at Engage
6. Google Search Console API: Provides Google visibility and performance data (whatever that’s worth). Yep, you guessed it: ScreamingFrog can connect to the API and sniff that data for you
7. ahrefs, SEMRush, or some other API: Gives you page-by-page earned media discoverability (yes, they provide more than SEO data)
8. Google Sheets or similar: Stores all this juicy stuff
9. AI: Grabs embeddings. This amazing article by ipullrank shows you how and why, and this great piece by Everett Sizemore shows you how if you don’t spend all your time hunched over a computer
10. A Google Sheet, or a ScreamingFrog javascript, or a database, calculates cosine similarity based on those embeddings (see previous links)
11. You: Provide a scoring system. You need to weight the different numbers you just grabbed to create a formula. See my Engage presentation, which includes a link to a content scoring algorithm Google Sheet
12. You: Provide content classifications. I usually use prune/roll up/revise/leave/investigate. To do that, I use the content percentile rank from the previous step
13. AI: Narrows the field. I may tell it to completely ignore content marked “prune,” for example. That’s a small step that reduces costs
14. AI: Now, finally, AI starts doing whatever magical auditing work you need. User intent? Yep. Applicable personas? Maybe. Check for errors? No, I wouldn’t do that. There are less-expensive APIs for that
15. AI: Assesses content using guidelines I put into my Claude Project, or my GPT, or my custom Gemini thingy. I may tell the AI to ignore all pages in the 10th percentile, or automatically mark all pages in the 95th percentile for revision
16. AI: Offers interlinking and rollup recommendations, although I haven’t had much luck there. I strongly recommend you check all recommendations before implementation

Note: If you want to take your audit to 11/10, have a look at DataForSEO, which is for a lot more than SEO. You can use their API to grab social media data, do sentiment analysis, etc. etc. It can take even more of the burden off of AI, reduce costs, and introduce more data into the mix. But at this writing, using their API will require a smidge of code.

What I didn’t use

Notice what I didn’t use?

• MCP servers. Love ’em. But I once tried to use them to grab data from Google Search Console. After two days of one bug after another, my head exploded. I got extensive therapy, then went back to APIs. MCP is not ready for real-world marketing applications. Plus, expensive
• Any code. At all. I didn’t write one. line. of. code. I’ll write another piece on the ups and downs of vibe coding. I’ve gotten some great results, but I’m not going to trust my clients’ content strategies to Python code I’ve written. Bad idea
• AI as a scraper. It’s expensive. It’s ineffective. It eats up a city block’s worth of electricity (might be a slight exaggeration, or a slight underestimation)

And, AI is, what, five out of sixteen steps? It’s improving things. But I’m not throwing my entire process in the trash.

Disclaimer about content generation

This post isn’t about content generation. I’ll unpack that barrel of raging caffeinated monkeys another time.

A note about specialized models

Britney Muller says it best: Specialized models are coming. They’ll be more efficient and a lot better at their assigned tasks. Yet another topic for another time.

That’s all I got

No fancy conclusions here, folks. Use AI to enhance your content auditing toolset, not replace it.