Why your traffic dropped after you used ChatGPT for all your content

The pitch is hard to resist: “Use ChatGPT to write 50 blog posts. Publish them all. Rank for hundreds of keywords. Scale content without hiring writers.”

I’ve watched at least a dozen Mumbai businesses try this in the last year, and it goes the same way every time.

Month 1–2: posts go live and some get indexed. A few start ranking for long-tail keywords, traffic ticks up, and everyone’s excited.

Month 3–4: Google crawls more of the site. The Helpful Content signal evaluates the site as a whole, and traffic plateaus.

Month 5–6: a core update or Helpful Content update rolls out. Traffic drops 40–70% and the posts that were ranking disappear. Worse, the pages that ranked before the AI content push also decline, because Google applies a site-wide quality assessment.

So the AI content didn’t simply fail to help. It dragged the whole site down.

Every “scale content with AI” enthusiast sees this graph eventually, and the drop looks like a cliff.

Why Google doesn’t penalise AI content (but penalises yours)

First, the biggest misconception: Google does not penalise content for being AI-generated. Google has said so explicitly. Content written by a human, by AI or by both is judged by the same quality standards.

What Google penalises is low-quality content, and most AI-only content is low quality. The grammar is usually fine. The problems are elsewhere:

1. It adds nothing new. ChatGPT generates content by synthesising what already exists, so your AI post about “SEO tips for small businesses” is a remix of the 10,000 articles already written on the topic. Google’s algorithm can tell that your page adds no original insight.

2. It lacks first-hand experience. Google’s E-E-A-T framework rewards Experience, meaning content from someone who has done the thing. AI can write about WordPress malware cleanup, but it can’t describe what the backdoor file looked like at 2am or how the client’s face changed when the site came back online.

3. It reads like everything else. AI has a recognisable style: hedging language, even-handed “on the other hand” structures, smooth transitions that say nothing. Readers (and Google’s classifiers) are getting better at spotting the sameness.

4. It has no author entity. AI content published without a real author (no name, no bio, no credentials, no LinkedIn) fails E-E-A-T at the most basic level. Google can’t verify expertise if there’s no person behind the page.

The number worth knowing: 86.5% of top-ranking pages contain some AI-assisted content. AI-assisted is a long way from AI-only, though. The top pages use AI for research, drafting and structure, and then a human expert adds original insight, edits for voice, and puts their name and reputation on it.

The Helpful Content Update: how it works

Google’s Helpful Content system works as a site-wide signal, so it doesn’t judge each page on its own.

If your site has 100 pages and 60 of them are thin AI content that adds nothing, the Helpful Content classifier may apply a negative signal to your entire site. Your 40 good pages get dragged down by the 60 bad ones.

That’s what makes the AI content blitz so dangerous. Publishing 50 AI posts risks more than those 50 posts not ranking; it can suppress everything else on your site.

Symptoms of a Helpful Content hit:

• Traffic declines across the whole site, including pages that aren’t AI-written
• Rankings drop for keywords you used to hold comfortably
• New content struggles to get indexed
• The timing lines up with a Google core or Helpful Content update

The AI content audit: diagnosing the damage

If you’ve published AI content at scale and your traffic dropped, assess it like this.

Step 1: Identify the AI content

Go through your blog and answer honestly for each post:

• Was this written entirely by AI with minimal editing?
• Does it contain original data, examples, or personal experience?
• Would a reader learn something they couldn’t get from any other page on this topic?
• Is there a real author with credentials attached?

Score each post: Keep (original, adds value), Rewrite (good topic, needs human depth), or Remove (generic, adds nothing).

Step 2: Check the timeline

In Google Search Console, compare traffic for:

• 3 months before the AI content was published
• The AI publishing period
• 3 months after

If there’s a clear decline starting during or after the AI publishing period, especially one that lines up with a Google update, the content is the likely cause.

Step 3: Assess the ratio

What percentage of your site’s content is AI-generated? Above 40%, Google’s Helpful Content classifier is more likely to flag your site, because the signal looks at the proportion of helpful to unhelpful content.

The recovery playbook

Option 1: Prune aggressively

For posts scored as “Remove”:

• Delete the page entirely
• Return a 410 (Gone) status code, not a 404
• Remove it from your sitemap
• Remove internal links pointing to it

The aim is to shift the ratio of helpful to unhelpful content. Go from 60% AI-generated to 20% and the site-wide signal improves.

Option 2: Rewrite with human expertise

For posts scored as “Rewrite”:

• Keep the URL (to preserve any existing authority)
• Rewrite with original insight, first-hand examples and expert opinion
• Add a real author bio with credentials
• Add schema markup (Article, Person)
• Include original data, screenshots or case studies AI couldn’t have generated
• Update the publish date to reflect the rewrite

When you’ve finished, ask: “Would someone in my industry share this as a useful resource?” If not, keep going.

Option 3: Consolidate

If you published 8 AI posts on variations of the same topic, merge them into one thorough guide:

• Pick the best URL
• Redirect the others to it (301)
• Rewrite the consolidated page with real depth
• Add original data, experience, and your perspective

The same consolidation approach works for ranking recovery after Google updates.

How to use AI for content without getting penalised

Avoiding AI entirely isn’t necessary. You need to use it properly, and this is the framework I use with clients.

AI does:

• Research and outline: “Give me the top 10 questions people ask about [topic]”
• First draft: “Write a 1,000-word draft covering these 5 points”
• Formatting: “Turn this into a comparison table” / “Rewrite this as bullet points”
• Variation: “Give me 5 different headline options for this post”

Human does:

• Strategy: decides what to write and why it matters for the business
• Original insight: adds personal experience, client examples, data from real work
• Voice: rewrites to match the brand tone and strips out AI hedging
• Fact-checking: verifies every claim (AI hallucinates, especially with numbers)
• Author attribution: a real name, bio and credentials attached to the post

The ratio that works

In my experience, content that ranks well in 2026 is roughly:

• 30% AI: research, outline, first draft structure
• 70% human: original insight, editing, voice, fact-checking, experience

Content that gets penalised is closer to:

• 90% AI: generated, lightly proofread, published
• 10% human: someone changed the title and hit publish

The 30/70 split is faster than writing everything by hand and still carries the quality and originality signals Google rewards.

The content that AI cannot write

However good the models get, some content needs a human in a specific situation:

• “I tried X and here’s what happened”: first-person experience with real outcomes
• “My client had this problem and we solved it by…”: case studies, with permission
• “The data from our 50 audits shows…”: original data from real work
• “Everyone says X but in my experience Y”: contrarian takes backed by practice
• “Here’s a screenshot of the actual dashboard…”: original material AI can’t produce

This is the content that builds E-E-A-T, earns links, gets cited in AI Overviews and keeps paying off for years. It works because it’s harder to produce.

The uncomfortable question

If your AI content got penalised, ask yourself honestly: would you have published it if a human writer had sent it to you?

If you’d have asked for a rewrite, AI wasn’t the problem. You lowered your quality bar because the content was free and fast.

AI produces what you ask for. Ask for “50 blog posts fast” and you get 50 thin posts. Ask for “a well-researched first draft on a topic I’ll add my expertise to” and you get useful raw material.

The businesses that got hurt published AI content because it was cheap. The ones that did well used AI to make better content than they could previously afford, and kept a human expert between the model and the publish button.

If your traffic has already dropped and you’re trying to work out how much is recoverable, that diagnosis is what I do. Recovery is a core service line, and what still works after AI Overviews is the strategy I rebuild sites onto. I give a fixed quote once I’ve seen the Search Console data, and if the honest answer is “start over on a subfolder,” I’ll say so.