The short version: Between mid-2025 and early 2026 my blog's Google traffic fell about 80 per cent, from roughly 4,700 clicks a month to 937 at the bottom. I rebuilt it with AI-assisted content published at volume, with a human editorial layer on top. Six months on, clicks are climbing again and my average Google position has gone from about 41 to 15. The honest lesson: AI content rebuilds traffic when you pair volume with indexation work and keep real editorial judgement. Volume on its own does not.
Almost nobody shows the real numbers when they write about AI content and SEO. So here are mine, month by month, straight from Google Search Console.
| Month | Clicks | Impressions | Avg position |
|---|---|---|---|
| Jul 2025 | 4,724 | 719,871 | 41 |
| Aug 2025 | 4,289 | 801,648 | 42 |
| Sep 2025 | 3,338 | 528,843 | 34 |
| Oct 2025 | 2,933 | 293,603 | 19 |
| Nov 2025 | 2,064 | 275,032 | 20 |
| Dec 2025 | 1,406 | 204,751 | 32 |
| Jan 2026 | 1,322 | 209,381 | 27 |
| Feb 2026 | 937 | 142,439 | 21 |
| Mar 2026 | 963 | 132,683 | 26 |
| Apr 2026 | 1,011 | 131,239 | 22 |
| May 2026 | 1,226 | 291,973 | 17 |
| Jun 2026 (to date) | 1,046 | 230,262 | 15 |
What happened
My traffic peaked in summer 2025 at about 4,700 clicks and 720,000 impressions a month. Then it fell off a cliff, the way a lot of content sites did through that period. By February 2026 I was down to 937 clicks. An 80 per cent drop is the kind of number that makes you want to close the laptop and not open it again.
Instead I rebuilt, in public, with AI. Not AI dumped raw onto the page, which is what gets sites penalised. AI-assisted content written against real keywords, with a human editorial layer, structured so search engines and AI engines can use it.
What worked
- Volume against real keywords, not random topics. Every post targets a keyword with real search demand, picked from data, not guessed.
- Fixing indexation, not just publishing. This was the biggest lever and the least glamorous. Publishing faster does nothing if Google will not index the pages.
- Ranking higher per query. My average position improved from about 41 to 15 across the whole site. The content ranks far better per query now than it did even at the old traffic peak.
- A human editorial layer. The line Google penalises is unhelpful content at scale, not whether AI was involved. Editorial judgement is what keeps you the right side of that line.
What did not work, and the honest hard part
Volume on its own. At one point I was publishing dozens of posts a day, and most of them never got indexed. The real ceiling was never how fast I could publish. It was how much Google was willing to crawl and index, and that is governed by site authority and quality, not output speed. If I could go back, I would have spent less energy on raw volume and more on indexation and authority from day one.
What this means if you are rebuilding too
AI content works for recovery. The data above is the proof. But it works as part of a system: real keyword targeting, an editorial layer, and relentless attention to whether pages are getting indexed at all. If you publish AI content at volume with none of that, you will get the penalty everyone warns about. If you publish it with all of that, you get a recovery curve like the one in that table.
To wire this into your day-to-day, see how to run AI marketing operations.
What I Stopped Publishing Once I Saw the Data
Three months into this rebuild, I pulled every AI-assisted post into a spreadsheet and sorted by organic sessions. The bottom 40 percent had one thing in common: they were listicles built around a keyword tool's suggestion rather than a question I'd heard from a client or in a Facebook group. "10 Social Media Tips for Small Business" pulled in 62 sessions over three months. A post answering "why did my LinkedIn engagement drop after I started posting daily" pulled 1,400 in the same window, despite having a much smaller estimated search volume in Ahrefs. I killed the entire listicle format after that and haven't written one since.
The pattern held across 34 posts I audited. Anything where I used AI to draft from a generic prompt like "write a blog post about email marketing best practices" underperformed. Anything where I fed the AI a transcript of an actual client call, a support ticket thread, or my own notes from a webinar Q&A, then asked it to structure and tighten that material, outperformed. The difference isn't the AI. It's what you feed it. I now keep a running Google Doc of every question a prospect or follower asks me and treat that as my only content brief source, AI included.
One number that surprised me: posts under 900 words that answered one specific question outperformed my 2,500 word "ultimate guides" on a per-hour-invested basis by roughly 3 to 1. The guides still rank and still bring traffic, so I'm not abandoning them, but I stopped treating length as a proxy for quality. My best performing piece from this whole run is 740 words and took under 90 minutes start to finish, including editing.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
- Cut: broad "tips" listicles with no source material behind them
- Kept: narrow questions pulled from real client interactions
- Changed: my AI brief now requires a pasted transcript or notes, never a topic alone
If you're rebuilding traffic with AI content and it's stalling, check whether your prompts start with a keyword or with a real conversation. That single swap explains more of my traffic recovery than any tool, plugin, or publishing schedule I tested.
Frequently asked questions
Does AI content hurt your Google rankings?
Only if it is unhelpful content produced at scale with no editorial layer. Google penalises unhelpful content, not the use of AI. AI-assisted content that is truly useful, targeted at real search demand, and edited by a human does not trigger those penalties. My average position improved from 41 to 15 while publishing AI-assisted content.
What matters more, publishing volume or indexation?
Indexation. Publishing speed is meaningless if the pages do not get indexed. The single biggest lever in my recovery was getting more of the existing pages indexed, by freeing crawl budget and building internal links and authority, not by publishing faster.
How long does it take to recover lost search traffic?
In my case the bottom was about five months after the peak decline began, and meaningful recovery took another four to six months of consistent work. Position improvements show first, then clicks follow as those positions climb onto page one.
If you want help building AI workflows into your own business the way I rebuilt this one, here is how to work with me.
For the bigger picture, see my full guide to AI marketing.
Related reading: How to Generate Traffic and Boost Your Blog's Following and How I Rebuilt a Newsletter Open Rate from 11 to 70 Percent.
Run a business, a brand or a blog? I feature tools, products and expert guest posts for an audience of 15,000. See how it works
