Yes, Search Console can automate a good part of your content work: collecting queries, sorting them and writing the brief. Choosing the topic and reviewing the draft can't be automated. At Pillet Grenié Bureau, a daily report shows us what people type before they land on our site. That report steers our next articles, and a person signs off at every step.
What your Search Console report really tells you about your next articles
Our own site analytics don't show us the query a visitor typed. Search Console and Bing Webmaster Tools are where we learn what people were looking for before they reached us.
Our first five days of imports gave us a few hundred impressions and three clicks. Almost every query was in French and about task automation: "automatisation des tâches", "automatisation tâches répétitives", "automatiser taches". Our average position sat around 50 to 65. That's pages 5 and 6. In our experience, very few searchers get that far, which fits the missing clicks.
Our reading is simple. Google already links us to the right topic, but our content doesn't rank yet. For an editorial plan, that's great news. We no longer have to guess topics. We just read them.
Bing sent a different signal. Only about ten impressions, but in positions 1 to 10, on long natural-language queries. They look like the searches Copilot or ChatGPT send to Bing to back up an answer. It's an early hint that AI assistants use our articles as sources. It isn't proof yet.
Why not just use the interface? According to Incremys, it shows 1,000 rows and 16 months of data, while the API returns up to 50,000 rows per request. If you want to cross pages and queries every week, store the history yourself.
Two traps to know about:
- Hidden queries. Clicks per query never add up to the total, because Google hides very rare searches that could identify a person.
- Noise. One consultant found prompt fragments pasted in as searches in his export. Not everything on the list deserves an article.
From report to draft: what we automated and what we keep by hand
Data collection: automated
Every morning, Google and Bing data lands in our own database, on Supabase, hosted in Paris. For each query we pull impressions, clicks, click-through rate and average position, broken down by page, country and device. In our imports, the latest Google figures arrive a few days late, so we re-import the last few days on every run to miss nothing. In our setup, Bing data comes in as weekly totals, so we read it week by week.
In plain terms, one line of the report looks like this:
query : automatisation des tâches
engine : Google
impressions : 55
clicks : 0
position : pages 5 to 6
A frequent query, zero clicks and a distant position: Google already gives us this topic, but we don't have a page strong enough to answer it yet.
The brief: automated
Each query goes into one of three buckets: new article, new section in an existing article, or discard. The brief then pulls together the target queries, the pages involved, the questions to cover and our notes from real projects. A first draft is written from that brief.
By hand
A person picks the topic from the suggestions. A person reviews the text before anything goes live. Those are the only two manual steps, and we won't give them up.
Why we never publish a generated article without review
We once ran a fully automatic content pipeline for our social media: planning, generation, formatting and publishing. We shut it down in October.
The first reason was topic drift. A machine follows whatever is easy to produce. When we audited our site, we found our topics leaned towards automation for solo founders. Not one article covered assessment, integration or visibility, the things we actually sell. A machine left to choose its own topics widens that gap instead of closing it.
The second reason: in our reviews, we've seen answer engines repeat a site's wording almost word for word. During a review for an independent gym, AI answers echoed the site's description almost verbatim. They named a discipline the gym didn't teach, while a competitor captured the answers for the ones it did. If a generated article states something false, that error becomes the answer your prospects read.
One SEO consultant recommends automating collection, checks and alerts, and leaving decisions on intent, evidence and publishing to humans. We agree. Our review checks three things: every claim matches reality, the example comes from a real project, and the answer shows up in the first few lines.
How we set this up for a client
We start with the data, not the writing. On day one, Bureau checks four things:
- Does a property already exist? A Google verification already on the domain means someone owns it, sometimes with months of history. Better to be added as a user: in our experience, a new property starts without that history.
- The right property type. On our own site, the first import failed: it was looking for a "Domain" property when only a "URL prefix" property existed. We caught it straight away because we check the first run by hand.
- Read-only access for the account that imports the data. It reads, it changes nothing.
- An alert if the import comes back empty. Without a valid access key, an engine is simply skipped, with no error. A report that stops silently is worse than no report: you think nothing is happening when the data is missing.
Before any new article, we also compare what the site claims with what the business actually offers.
What the client gets: every week, they see the questions Google associates them with and the pages that get seen but not clicked, then approve one or two briefs. Their team spends its time deciding, not exporting spreadsheets. That's the heart of our integration and automation work.
What results to expect, and how soon
Let's be honest: nothing moves in a week. In our experience, a new property starts with no history, and Google's latest numbers arrive a few days late. It takes us several weeks to read trends, and we plan for several months before positions change.
Watch the signals in this order:
- Impressions on your target queries.
- Position, moving from page 5 towards pages 1 and 2.
- Clicks.
- Contact requests.
Google itself presents the queries map in Insights as a way to spot content to create or improve.
For AI assistants, visibility is still partial. Search Console's generative AI report only shows impressions for now, without the queries. The long queries we see in Bing are our best early indicator so far.
Our own starting point is on record: pages 5 and 6. We'll measure what happens next with the same method we use for clients and publish the before and after. The machine spots the topics. A person decides what gets published.
