Guides & field notes
The guides explain how to use Atelier SEO. Field articles start with a concrete problem, measure it and show how to address it.
How-to guides
What an audit shows, how to pick your projects, what Iris can do, and what it won't.
In June 2026, Google opened a Search Console report on the generative AI features of Search. It counts impressions in AI Overviews and AI Mode, and nothing else: no clicks, no queries, and a gradual rollout. ChatGPT and Perplexity open no console at all. Here is what a survey done by hand still proves.
Three projects compete for the same budget, and there is no universal order between them. There is a reading order: you go through the signals in a set direction and stop at the first one that blocks. That stopping point is your priority, and it is not the same from one site to the next.
A site audit is only worth its order. Three layers, three decision thresholds, and the one check that classic audit grids leave out.
A single reading gives a snapshot, not a trend. The cadence, the reading grid and the decision rule that turn a one-off test into AI visibility tracking.
The number of clicks between a page and the home page decides how often Google crawls it. How to measure it, what threshold to keep, and what to fix first.
Improve, merge, archive or keep: the four outcomes for a page with no traffic, the trigger behind each one, and how to sort the list before starting.
In the field
Internal linking, visibility in answer engines, performance that won't budge.
A basket of ten questions, asked every month in private browsing on ChatGPT and Perplexity, to find out whether your company is quoted by AI assistants. The protocol, the reading grid, and the observatory applied to this site.
A tool is enough in some cases, not in others. Five focused questions help you choose between self-service, hands-on work on your site, or a full redesign — without pushing the most expensive option.
A website redesign rarely fails because of the design. It fails because of what wasn't checked beforehand: the pages that drive results, the redirects, the Google exports, and the date of the first benchmark.
Our method
Every article starts with a problem observed in an audit or during client work: an important page that does not rank, internal links that scatter authority, or loading time that resists the first optimisations. We begin with the measurement that confirms the problem, then examine plausible causes and the checks that separate them. A recommendation without evidence or a success criterion is still an opinion, so it does not belong in our method.
We then rank actions by expected impact, effort, dependencies and risk. Fixing everything a tool reports is rarely the right strategy: some warnings are informational, others describe the same defect from several angles, and a few depend on an external service. The aim is to choose the next project that matters to the business, explain that choice, and preserve a measurable baseline before anything changes.
After publication, we return to the data: a fresh crawl, a comparison of the metrics, a check of the affected pages, and traffic observation once enough time has passed. If the result does not move, the article says so and the approach changes. These guides are therefore designed both to apply the audit’s recommendations and to understand their limits, without confusing a score, search visibility and a commercial outcome.
Apply it to your website
First verify the findings on your own domain, review our public tracking page to see the method in practice, then compare support levels if you want to delegate implementation.