AI Content Audit: How to Spot AI Slop on Your Own Site

15+ years in local marketing; Google Ads certified; Shopify Partner.

Here's a scenario I see constantly. A business owner asks me to look at their site. I open it up, and within about 90 seconds I can tell they've been publishing AI-generated content, probably for the last year or two, probably without realizing how obvious it is.

Not obvious in a "this is clearly a robot" way. Obvious in a "this says nothing, helps nobody, and Google knows it" way.

That's what an AI content audit is really about. Not catching AI for the sake of catching it. Catching the content that's quietly dragging your site down, whether a human wrote it badly or a machine wrote it fast.

This article is the practical guide. I'll show you what to look for, how to check it yourself, and what to actually do about it. For the broader question of how AI fits into your marketing overall, our complete guide to AI for marketing covers that ground. This article is specifically about auditing what's already on your site.

What "AI Slop" Actually Looks Like (And Why It's Hurting You)

Let me be honest about something first. The problem isn't AI-generated content. I use AI in content production all the time. The problem is AI-generated content that was published without editing, without a real point of view, and without any connection to what your actual customers need to know.

I call it slop because that's what it is. It's content that fills space without earning it.

Here's what slop looks like in practice.

The generic opener. "In today's competitive landscape, businesses need to stay ahead of the curve." That sentence means nothing. It could describe any business in any industry in any year. If your blog posts open like this, you've got slop.

The listicle that teaches nothing. "5 Reasons Why [Your Service] Matters." Each point is one sentence. No examples, no specifics, no real explanation. A reader finishes and knows exactly as much as when they started.

The keyword-stuffed paragraph. The service is mentioned seven times in four sentences. It reads like the writer was being paid per mention. Because the AI was, sort of, it was told to optimize for a keyword and it did, badly.

The confident-but-wrong claim. This one is dangerous. AI tools hallucinate. They state things with total confidence that are just wrong. In regulated industries like law, healthcare, or financial services, that's not just embarrassing, it's a liability.

In my experience, sites that published AI content aggressively in 2023 and 2024 without editorial review are now sitting on a backlog of pages that either rank for nothing or rank for terms that don't convert. The content exists. It just doesn't work.

How to Run a Basic AI Content Audit Yourself

You don't need a fancy tool to start. You need a spreadsheet, a few hours, and honest eyes.

Here's the week-by-week process I'd walk you through.

Week 1: Pull your content inventory.

Go into Google Search Console (it's free, and if you don't have it set up, stop reading and set it up right now). Export your top 100 pages by impressions over the last 12 months. That's your starting list.

Add a second list: any page published in the last 18 months. You can pull this from your CMS. Sort by publish date.

Combine both lists in a spreadsheet. You're looking for overlap and gaps. Pages that get impressions but no clicks are a red flag. Pages with decent clicks but zero conversions are a different kind of problem.

Week 2: Read the pages out loud.

I know that sounds low-tech. Do it anyway.

Read the first paragraph of each page out loud. If you stumble over it, or if it sounds like a press release from a company that doesn't exist, flag it. If you get to the end of a paragraph and can't summarize what it just said, flag it.

Look for these specific signals:

Sentences that start with "It is important to note that..."
Paragraphs that end with a vague transition like "With all of this in mind, let's explore..."
Headers that promise specifics and body copy that delivers generalities
Any claim you can't verify or attribute to a real source

Week 3: Run the numbers on each flagged page.

For every page you flagged, pull these four numbers from Search Console and Google Analytics:

Impressions (how often it showed up in search)
Clicks (how often someone actually visited)
Average position (where it ranks)
Bounce rate or engagement rate (did they stay or leave immediately)

A page that ranks in position 8-15, gets impressions, but has a click-through rate under 2% often has a title and meta description that's too generic. That's usually an AI tell, the title says "Complete Guide to [Topic]" and gives no reason to click.

A page that ranks, gets clicks, but has an engagement rate under 30 seconds is worse. People showed up and immediately decided this wasn't worth reading.

Week 4: Prioritize and decide.

Now you have a real list. Sort it into three buckets.

Bucket 1: Fix it. Good topic, decent ranking, fixable problems. These get rewritten or heavily edited.

Bucket 2: Merge it. Thin content on a topic you cover better elsewhere. Redirect these pages to the stronger version.

Bucket 3: Delete it. No rankings, no traffic, no purpose. Gone.

Most sites I audit end up with about 20-30% of their content in Bucket 3. That's not a failure. That's clarity.

The Specific Signals That Tell Me a Page Was AI-Generated (Without Checking a Tool)

I've read enough AI content now that I can spot it without running it through a detector. Here's what I actually look for.

Structural tells.

AI defaults to a specific content shape: intro paragraph, three to five headers, bullet points under each header, conclusion paragraph that restates the intro. It's not always wrong. But when every single page on your site follows this exact shape, it's a tell.

Real content has variation. Some points deserve a long explanation. Some deserve one sentence. AI doesn't make that judgment call, it fills the template.

The authority-without-specifics problem.

AI writes confidently. It sounds like an expert. But it rarely gives you a specific number, a specific example, or a specific situation where the advice applies.

Compare these two sentences.

"Google rewards websites that publish high-quality content regularly."

"Sites that publish one well-researched 1,500-word article per month consistently outperform sites that publish eight thin 300-word posts, per Ahrefs' content study data."

The first one is technically true and completely useless. The second one gives you something to work with. AI almost always writes the first version.

The missing "so what."

Good content answers a question and then tells you what to do with the answer. AI often stops at the answer. It explains what something is, but not what you should do about it, or why it matters to your specific situation.

If your pages read like encyclopedia entries, that's the problem. Useful content is opinionated. It says "here's what I'd do" or "here's what this means for you." AI hedges constantly because it's been trained to avoid being wrong.

The fake expertise tell.

This is the one that gets businesses in trouble. AI will write about your industry using the right words, but it doesn't know the difference between what's technically accurate and what's actually true in practice.

I've seen dental sites with AI content that described procedures incorrectly. Legal sites that cited outdated case law. Trades businesses with safety information that was just wrong. The content passed a surface-level read. It didn't pass a professional one.

If you're in a regulated industry, this matters a lot. Under Canada's Competition Act, misleading advertising claims, including false technical claims on your website, can attract real regulatory attention. That's not a hypothetical.

What the Numbers Actually Tell You About AI Content's Performance

Here's where I want to give you something concrete, because a lot of the conversation around AI content is vibes-based. Let me give you math instead.

Per DataForSEO's Canadian keyword data, the average CPC for "ai seo" in Canada is CA$21.33, and for "ai seo tools" it's CA$28.58. That tells you something important: the clicks you'd have to pay for in Google Ads are expensive. Organic content that earns those clicks is worth real money.

Now think about what happens when a page ranks for a keyword but doesn't convert. Say you rank position 6 for a term with 1,000 monthly searches in Canada. Position 6 earns roughly 3-4% click-through rate, so call it 35 clicks a month. If your content is slop, and those 35 people bounce in under 30 seconds, you've earned nothing. But if you'd had to pay for those 35 clicks via Google Ads at CA$21/click, that's CA$735/month in traffic you're wasting.

Multiply that across 20 underperforming pages and you're looking at a meaningful number. Not because I made it up, but because you can do this math yourself with your own Search Console data and the CPC benchmarks from DataForSEO.

That's the real cost of slop. Not Google penalizing you (though that happens too). It's the compounding waste of traffic that arrives and immediately leaves.

Per a 2024 BDC study, 66% of Canadian entrepreneurs reported using AI in some form, including many who weren't aware they were doing it. That means a lot of Canadian SMB sites have AI-assisted content on them right now, some of it good, a lot of it not. The ones who audit it and fix it are going to be in much better shape as AI search engines like Perplexity and Google's AI Overviews get better at recognizing thin content.

On that note: if you want to understand how AI Overviews affect what shows up in search results, our breakdown of how to rank in Google AI Overviews is worth reading after this. And if you're thinking about how AI search engines like ChatGPT and Perplexity find and cite your content, earning AI citations is the next logical question.

The Difference Between "AI-Assisted" and "AI-Generated Slop" (This Is the Piece)

I want to be clear about something, because I don't want this article to read as "AI bad, humans good." That's not what I think.

AI-assisted content, where a human uses AI to research, draft, or structure, and then edits it heavily with real expertise and real opinions, can be excellent. I've seen it work well. Our own content process at Unalike uses AI tools at certain stages.

The slop problem comes from a specific workflow: prompt AI, copy output, publish. No editing. No fact-checking. No added perspective. No "here's what this means for someone in Regina running a trades business."

Here's how I'd describe the difference in practice.

AI-assisted content sounds like a person who did their research and has an opinion. It has specific examples. It makes a recommendation. It might say "I've seen this go wrong when..." because there's a human behind it who has actually seen things go wrong.

AI-generated slop sounds like a Wikipedia article written by a committee that's never met. It's accurate in a general sense. It's useless in a specific sense.

The audit question isn't "did AI touch this content?" The audit question is "does this content actually help a real person make a real decision?"

If the answer is no, it doesn't matter who wrote it.

For the tactical side of producing AI content that actually works, our guide to AI content writing for SMBs goes into the production process in detail. And if you're thinking about the SEO implications of how AI reads and processes your content, LLM SEO covers the technical side of that.

What to Do With the Pages You Flag

Once you've got your three buckets from the audit, here's how I'd approach each one.

Fixing a page means adding what AI left out. Specifics. Examples. Your actual opinion. A real recommendation. If the page is about "how to choose a [service provider]," it should say what you'd actually look for, not a generic list of considerations.

Add a number somewhere in the first 200 words. Real numbers, cited if possible, are one of the fastest ways to signal to both readers and search engines that a human with real knowledge wrote this.

Merging pages is simpler than it sounds. Pick the stronger page, redirect the weaker one to it using a 301 redirect (your web developer can do this in five minutes, or you can do it yourself in most CMS platforms). Update the stronger page to cover any unique angle the weaker one had.

Deleting pages requires one step beyond just hitting delete: set up a 301 redirect from the deleted URL to either your homepage or the most relevant existing page. Don't leave dead URLs. They confuse both visitors and search crawlers.

One pattern I see consistently across sites: businesses that clean up 30-40% of their content and redirect it properly almost always see the remaining pages perform better within 60-90 days. Not because Google rewards you for deleting pages, but because your good content stops competing with your bad content for the same keywords.

When to Do This Yourself vs. When to Hire Someone

Honest answer: the audit itself, you can do. The execution is where it gets complicated.

If you've got under 50 pages on your site and a few hours to spend, the process I described above is something you can run yourself. Pull the data, read the pages, flag the problems, prioritize.

If you've got 200+ pages, multiple service areas, and content that's been accumulating for years, you probably want help. Not because the process is different, but because the volume is unmanageable solo, and the prioritization decisions get more complicated when there are a lot of interdependencies.

What you should never pay for is an "AI content audit" that's just a report. A report that tells you your content has problems isn't worth anything. You need the fix, not the diagnosis.

If someone is pitching you an AI audit as a standalone deliverable with no clear path to implementation, that's worth being skeptical about. For a fuller look at what AI readiness audits actually include (and what they cost), this breakdown of AI readiness audits is a good reference. And if you're evaluating an AI marketing agency to help with this work, that guide covers what to ask and what to pay.

3 Takeaways

1. The problem isn't AI, it's publishing without editing. Content that doesn't help a real person make a real decision is slop, regardless of who or what produced it. An AI content audit is really a usefulness audit.

2. The audit process is four weeks of honest work. Pull your data from Search Console, read your pages out loud, flag the generic ones, run the numbers, then sort into fix/merge/delete. You can do this yourself for a smaller site.

3. The cost of slop is real and calculable. Traffic that arrives and immediately leaves has a dollar value. Multiply your underperforming page traffic by the CPC equivalent and you'll see what bad content is actually costing you.