I came into this interview with a number I could not explain. One query. Ten searches a month in the keyword tools. Eleven hundred impressions for that same query in Google Search Console, on my own site, over the same month. I had published that gap across seven platforms the day before we recorded. Then I handed it to Gert Mellak, who has run SEO out of Madrid for twenty years and worked with 500+ businesses, and asked him what I was missing.
He came back to it about ten minutes later with the answer.
“We know now that Search Console is heavily impacted by bots. Lots of bots produce impressions. I used to use impressions all the time. A few months ago, still, I talked about impressions, and look, we are more visible, etc. All crap. Because we know bots are creating impressions and Google doesn’t care to filter them, apparently.”
Gert Mellak
I had arrived at the same conclusion from the other direction and had nobody else to check it against. Now a working consultant with two decades behind him has said it on tape.
The query that started it
Here is the case exactly as I put it to him. A search query with a listed volume of ten searches per month in the US. That volume came from OpenSEO, which pulls DataForSEO, which pulls the Google Ads API and adds no extra layer on top of it. Ahrefs adds its own layer and moves those numbers up or down. So the ten was about as close to Google’s own number as I could get.
Same query. Same month. My site. Eleven hundred impressions in Search Console.
Both numbers claim to describe demand for one string of words. They are off by a factor of a hundred and ten. When I pulled the daily export, the shape gave it away: 52% of those impressions landed inside three days in July 2025, at an average position of 72, with zero clicks. No human behaves like that. I wrote the longer version of why Google Search Console data is not accurate before this interview. Gert got there from client work instead of from a spreadsheet.
He was reporting impressions a few months ago
The part that stuck with me is that this is a recent change for him. “I used to use impressions all the time,” he said. “A few months ago, still, I talked about impressions, and look, we are more visible, etc.” Then he stopped, because the number stopped describing anything.
He added a wrinkle most reporting has not caught up to yet. “We even see now that AI mode queries also show up in Google Search Console,” he said, “so somebody asking a follow-up question is going to produce another query.” One person asking three follow-ups produces three query lines. Good to know it is in there. Bad if you were treating the count as a headcount.
The num=100 change already told us this
When Google removed the ability to pull a hundred results per page, half my impressions disappeared overnight. I said on the episode what that has to mean going the other way:
“That means every search volume that we’ve seen, that’s a ton of scraping, especially for money keywords. And the more money that those keywords have, the more scraped they are, the higher the search volume.”
Jeremy Rivera
The scrapers were rank trackers. They were generating impressions on our behalf for years, and that same scraping activity fed the volume estimates we then used to justify budgets. Google has no reason to correct it either, because a bigger search volume number sells more ads.
It shows up on the click side too. Before we hit record I mentioned a case where Google Analytics reported a hundred clicks total and fifty of them fell into direct. Microsoft Clarity was able to clarify that those were bot clicks. Half of a “direct” channel was machines.
Search volume has very little value on its own
Gert’s position on search volume is blunter than mine was going in. “I think search volume has really very little value as a metric on its own,” he said, and then he walked through why the math built on top of it collapses.
The standard model is volume times click-through rate times conversion rate times lifetime value. Every SEO has built that spreadsheet. Gert points at Rand Fishkin’s number as the thing that broke it: sixty-eight percent of Google searches do not lead to a click on a website.
“Thinking that you still can calculate an ROI based on a traffic metric or a search volume metric, I think is insane.”
Gert Mellak
He also has clients rejecting low-volume keywords for exactly the wrong reason. Content built around a ten-per-month query can connect a brand to a concept well enough to get recommended inside ChatGPT, and you will never see that in a click. Though a mention is not always a win. He ran a test on the top CRMs and Salesforce came up negatively: “They said, whatever you do, don’t go with Salesforce, because it’s absolutely not a fit here.”
The conversation nobody wants to have with the C-suite
This is the point where he challenged twenty years of my working method, and I had to sit with it for a second on air.
“The main challenge we have right now is to explain to C-level executives that the ROI on SEO is not measurable on its own. Unless you make a hundred and fifty assumptions that are all flawed on its own, coming with data from platforms that are all flawed on their own, and you accept this as the source of truth.”
Gert Mellak
The damage runs backwards into keyword selection. He described a client who picked broader keywords than their niche warranted, because those were the ones with enough volume to make the projection model return a positive number:
“We go for keywords in SEO that we wouldn’t go for, just in order to make sure the spreadsheet shows a positive expected ROI. Which is not going to happen after six months, because the keywords are useless.”
Gert Mellak
His fix when a client will not let go of a keyword is the cleanest thing in the episode. Buy it. Pick the three keywords you think will make or break the business, run ads on them for three weeks, then restart the SEO project. Clients come back saying they got zero conversions from three weeks of people searching that exact phrase, or that the 2,500 volume was really 120. A small ad budget settles an argument that would otherwise burn six months of retainer. It belongs on the same short list as the questions worth asking before you hire an SEO company, because both are cheap ways to find out early whether the plan holds up.
Gert’s three-week ads test is written up as a run-it-this-week procedure: SEO SOP: Validate Keywords With Three Weeks of Google Ads. Numbered steps, the decision rule, and a one-page PDF free inside The Vault.
I told him on the episode I was going to declare victory for him in the social post afterwards. He wins for losing.
What he tracks instead
Gert’s replacement metric is internal and boring in a good way. He calls it high intent, and he defines it as a soft conversion: people landing on the pages he expects them to hit before they convert. The service page. The course page. The pricing and features page for a SaaS product.
“It just tells me how many people are now getting to your key pages,” he said. It works across channels, which matters because SEO does not run in isolation and branded traffic gets attributed in every direction. If more people are reaching the key pages this month than last month, something is working, and you can trace backwards to find out what.
He is equally strict about scope. His team runs a master tracking list of maybe ten or twenty keywords at a time, not five hundred. Even on a site with 1,500 pages. And he refuses to report single rankings, because it makes him look like a liar: “There is no value for the client to pick one query and say, look, this is now ranking number three. They’re going to type it in, they’re going to see it on number five.”
From 1,600 articles down to five
The funnel he described for a client with fifteen years of content is the most reusable thing he gave away. Sixteen hundred articles. What do you actually want to achieve? Now it is 150. Who is your client? Now it is 13. What has been working in Google Ads? Now it is five.
Then you go deep on those five, look at what competitors are doing with the same topics, and build links to strengthen them. The other 1,595 get a decision later.
He also asks every client what data they are sitting on and never look at. That is where the feedback forms come in.
“We have clients who come with 5,000 feedback forms filled in, where clients put half a page of description why they are actually reaching out, what their problem is, what their pain point is, how they talk about it. For us, this is gold. And clients are about to delete it because it takes up so much space on their computer.”
Gert Mellak
Do SEO, but search engines don’t exist
His framework starts from a thought experiment. “Imagine you do SEO, but search engines don’t exist.” No Google. You still have to get the site in front of the right audience. What do you do?
You start with audience platform research, because you need to know where they are and what they talk about. He uses tools like SparkToro for that part. Then you set alerts so you can contribute to those conversations. Then you go after other people’s audiences, which is usually industry publications. Then you build the distribution before you write anything, through collaborations and partners and podcasts. Only then do you produce content, because now you know the topic and you know how it will travel.
He wrote this up as the Search Everywhere Optimization Pyramid, and the full version is live on Search Engine Land and on his blog.
“The funny thing is that if you do it this way, now bring search engines back. You have done exactly what you need to do in 2026 in order to rank well.”
Gert Mellak
The join-the-dots cat
He explains site architecture to clients with a children’s puzzle. Connect dot one, dot two, dot three, and it turns into a cat.
“If there are not enough points on the map, the cat doesn’t have a tail, or only has two legs. Not recognizable as a cat, right? You put too many points on a map, you can’t join them anymore. It’s too complex.”
Gert Mellak
He had a client whose content had so many unrelated points that Google could not resolve the picture. “We removed 80% and they started ranking for their core keywords again.” Three months ago he did it again with a consultant sitting on twenty years of content. They cut at least half. She is back on page one for her top keywords.
SEO is one wheel
The line that reframes the whole job came near the end. “Nobody cares about SEO,” he said. “I haven’t had a single client stay with me because the SEO was so good.”
“Nobody sells a trip to Hawaii by telling you how good the plane is, and how easy the check-in is, and how nice the flight attendants are, and how great your suitcase is suited for this trip. But this is what we do. We say, hey, you should go to Hawaii because we have the best plane.”
Gert Mellak
SEO is one wheel in the machine that gets the brand to Hawaii. An important one. Still a wheel. And a wheel does not need the absolute truth from its instruments, it needs enough of a directional read to know which way to turn.
That connects back to something Matt Brooks of SEOteric says, which I brought up on the episode: ChatGPT is a terrible customer support representative out of the box until you train it. Building the distribution network, doing the research, doing the off-page work, building an authority hub. That is how you feed the bots the right way instead of producing AI slop. Brett Schklar, my previous guest, put the other half of it well: AI’s not taking your job, it’s taking your tasks.
Where to find Gert
Straight from him at the end of the episode: “The best would be to head over to gertmellak.com, or find me with my name on social media. Mostly I spend time on LinkedIn and Instagram these days.” He is at gertmellak.com and on LinkedIn. His consulting work is at SEOLeverage, his AI practice is Solicom AI Consulting, and the attribution rabbit hole he said was “another episode” is already written up at precise attribution is a myth.
Full episode and transcript are on the site. If you take one thing out of it, take the impressions column out of your client report this month.
