The SEO Turtle journal
What we’re working on, what we’re learning, and the questions worth spending a little more time with.
Most redesigns that lose traffic fail for one of a handful of reasons, and every one of them is checkable before launch. Here is the order we work through, why mass-redirecting to the homepage is worse than a 404, and what changes now that ChatGPT and Perplexity crawl on their own schedule.
A crawl tool tells you what a bot could do. A server log tells you what it did. Most log file guides are written for sites with millions of URLs, then handed to people with four hundred pages and no SSH access. Here is the honest version: why crawl budget is probably not your problem, why the bot population is, and where the logs actually live on Cloudflare, Vercel, cPanel and Shopify.
Google has looked at the page and decided not to keep it. Most advice treats that as a technical fault and sends you off to hammer Request Indexing. Here is the order that actually finds the cause: an afternoon of cheap plumbing checks, a look at what Google chose as the canonical, and then the harder question of whether the page deserves to exist.
Almost everyone appeals first and fixes second, which burns the one clean appeal Google gives you. Here is the order that works: diagnose the suspension type, correct the actual breach, assemble the documents, then open the tool. Plus the appeal and Digital Services Act routes that Cyprus and UK businesses have and most guides never mention.
In the last week of August, Google started expanding AI Overviews into full AI Mode responses with the follow-up box already open. Days later, Search Console finished rolling out a report that lumps both surfaces into a single impressions figure. The distinction the industry spent eighteen months drawing has collapsed from both ends at once.
Google now routes search result links through google.com/goto, and the encoded URLs cannot be read without following every redirect. Rank trackers just got far more expensive to run, which means your rankings report is quietly degrading. Here is what we are telling clients to measure instead.
Google stopped applying site reputation abuse manual actions for people searching inside the EEA. The headlines called it a retreat. It is closer to a swap: the penalty goes, the algorithmic separation stays, and Google finally published the four criteria it judges this on. That last part is the useful bit, and it applies everywhere.
A synthetic voice rings your shop, asks what a brake inspection costs, and emails the answer to a customer who is comparing you against eight competitors. Most advice about this feature is about training staff to answer the phone faster. We think that misses the point entirely. The call is Google's fallback for data it could not already read.
Apple's published Applebot address pool went from 2,400 IPs to 7,056 with no blog post and no explanation, weeks before the rebuilt Siri ships in iOS 27. Most sites have never looked at how they treat Applebot, and a lot of them are blocking the wrong user agent. Here is what changed and what to check this week.
Most local businesses picked their Google Business Profile category once, at signup, and never looked at it again. A new study across 1.8 million profiles suggests that single dropdown is worth more positions than most of the content work sitting in your backlog. Here is what the data says, where it stops being true, and what to change this week.
The August 2026 spam update has a start date, an end date and one confirmed exclusion. The volatility that actually moved most sites happened five days earlier and Google never confirmed it. Attributing the wrong drop to the wrong event is how businesses end up fixing something that was never broken.
Ask Gemini the same local question twice and the sources it cites match roughly 40% of the time. The top recommended business matches about 8% of the time. That single fact undoes most of how businesses currently check their AI visibility, and it changes what you should actually be optimising for.
Reddit went from the most cited domain in ChatGPT Search to a rounding error in about a week, and no one at OpenAI announced anything. If your AI visibility strategy depends on being quoted from somebody else's platform, this is your warning shot.
New research pulled apart how ChatGPT actually fetches web pages, and the answer is uncomfortable. It runs its own index that barely overlaps with Bing, serves most answers from a cached copy of your page, and only truly opens about one page in eighty. Here is what that means for how you write and structure a page.
Businesses are buying GEO tools to fix problems a 2019 site migration created. AI search did not add new technical requirements, it just stopped compensating for the old ones. Here is our practitioner take on why the technical SEO backlog is now the AI visibility roadmap, and why that favours smaller sites.
One number has done more to move marketing budgets in the last year than any Google update: AI search traffic converts 23x better than organic. It came from a single site over 30 days, and the company that published it said so at the time. Bigger datasets tell a very different story.
Your AI visibility tool says you appear in thirty categories. Your sales pipeline says two of them are real. New research from Search Engine Land explains the gap: AI engines will cite you across almost any topic, but they only recommend you inside the category you actually own. That changes what a content plan should look like.
Everyone assumed AI search would get smarter at reading nuanced content. The models Google actually puts in front of your pages are getting lighter, because at Google's query volume cost wins. That flips the content advice: stop writing for a reader that infers.