Talk to Franz

SEO · 08

Search Data and Traffic Forensics Consulting.

Underwater, a woman at a wooden workbench ties a green ribbon around one glass jar on a brass stand, next to an open crate full of closed jars, each holding a folded note
AI-generated image (OpenAI, via Codex) for Full Stack Optimization

When traffic drops, everybody has a theory. I take your raw Search Console data, test one hypothesis at a time, and tell you which cause is proven and which is not.

You hire me when

  • Traffic dropped, and nobody can prove why.
  • A report blames a core update, and nobody in house can check it.
  • Search Console, analytics and your tools disagree.
  • You want to know what AI search really sends you.

What you get

  • Raw Search Console data via the API: 16 months, per page, query, country and day.
  • Every explanation as a hypothesis with a test and a verdict: proven, ruled out or open.
  • Changes dated from the data, next to your release log.
  • Brand vs non-brand, page clusters, year over year.
  • Analytics and consent tracking checked down to the request.
  • A findings workshop, and the scripts, so your team can re-run every number.

Why traffic losses get misread

A traffic drop has many possible causes at once: a release, a Google update, a change in demand, a change in how Google counts. Each team in the room sees the one it knows best.

Google lists the causes it wants you to check first, and one of them is not on your site at all: “The drop might be related to a change in the data processing or a logging error.” (Google Search Central, Debugging drops in Google Search traffic)

So I do not start with a theory. I start with the raw clicks. My goal: every explanation becomes a hypothesis with a test, and you get a verdict for each one: proven, ruled out, or still open.

Where traffic analyses go wrong in 2026

Most traffic reports I review have the data. They read it in the wrong order. These are the mistakes I found in my 2026 analyses, and what I specify instead:

What breaksWhat I measuredWhat I specify
The report measures a model, not your clicksThe named cause explained about 0 percent of the lossClicks from Search Console, every row, before any verdict
A counting change looks like a traffic lossImpressions almost halved, clicks down about a tenthClicks first, Google’s data anomalies page second
Nobody knows what shipped whenA release log with a gap of months right before the dropChanges dated from the data, next to deploy dates
A proven defect is not a proven causeA crash in 4 in 10 stored pages, its click share not measurableMechanism and share, proven separately
Traffic outside every listA third of clicks on pages in no sitemap; 0 impressions for paginated pages in 16 monthsClicked URLs, sitemaps and crawl side by side
Brand clicks hide the trendAbout two thirds of one site’s Google clicks were brandedBrand and non-brand as two curves
AI traffic read from headlinesAI referrals between 0.01 and about 1 percent across about 40 sitesMeasure it per source before you buy a tool

The report measures a model, not your clicks

An agency analysis blamed a traffic loss on a change to the indexing rules for one page type. It measured a visibility index.

  • I tested the claim on every row of the Search Console data.
  • The change was real. While it ran, it cost at most about 0.3 percent of clicks.
  • Its share of the loss: about 0 percent. It was the state before the loss, not its cause.

The full case is on my SEO agency cleanup page.

What I specify: clicks from Search Console are the primary metric. No verdict before every row is checked. Tool scores are simulations.

A counting change looks like a traffic loss

In one 2026 case, impressions almost halved from one period to the next. Clicks fell by about a tenth. The average position went from about 5 to about 9.

That shape does not say “we lost a tenth of our visitors” and it does not say “we lost half”. It says impressions and clicks must be read apart.

Google itself had a counting problem for almost a year: “A logging error prevented Search Console from accurately reporting impressions from May 13, 2025 until April 27, 2026” and “clicks were not affected by the error” (Search Console Help, Data anomalies in Search Console).

What I specify: clicks first, impressions second, average position last. Before any verdict, check Google’s data anomalies page for the dates in question.

Nobody knows what shipped when

In one 2026 analysis, traffic fell by double digits year over year. The release log had a gap of months right before the drop. Its dates were ticket dates, not deploy dates.

  • Two changes went live days apart, and both were followed by click breaks.
  • In one agency review, several production changes named in the agency’s reports were missing from the release log. I dated one of them from the data, to within days.

What I specify: one release log for every change that touches search, with the deploy date and an owner. Where the log is missing, the change gets dated from the click data, URL group by URL group.

A proven defect is not a proven cause

On one site, the JavaScript crashed in Google’s stored copy of 4 in 10 top pages. Google’s own test showed one of them with its main content gone. The defect was proven in the code and in Google’s renderer.

How many clicks it cost could not be measured. So I did not print a number for it. The details are on my technical SEO page.

What I specify: two separate answers for every finding. Is the mechanism real? And what share of the loss does it explain? A finding without the second answer is a risk to fix, not the cause of your drop.

Traffic outside every list

  • In one 2026 audit, pages that earned a third of all clicks were in no sitemap at all. About 40 percent of all clicks went to language versions of pages, also in no sitemap.
  • On another site, paginated pages had 0 impressions in 16 months. Google called page 2 “URL is unknown to Google”.

What I specify: three lists side by side: the URLs that get clicks in Search Console, the URLs in your sitemaps, and the URLs a crawl finds. Every URL that is in one list and missing from another is a finding.

Brand clicks hide the trend

On one 2026 site, about two thirds of the Google clicks were branded: people searching for the company’s own name.

A steady total can hide a falling non-brand curve. A brand campaign can hide an SEO problem for a year.

What I specify: brand and non-brand as two separate curves, year over year, before anybody reads the total.

AI traffic read from headlines

For my 2026 conference keynote I measured AI referrals across about 40 sites in their own analytics and Search Console data. AI assistants sent between 0.01 and about 1 percent of their traffic.

Google counts its own AI features inside web search: “sites appearing in AI features (such as AI Overviews and AI Mode) are included in the overall search traffic in Search Console” (Google Search Central, AI features and your website).

What I specify: AI referrals per source, from your own data, next to Google’s clicks. Then decide what a tool or a project is worth.

How I work

  1. Raw data first. Search Console through the API, per page, query, country, device and day, for the full 16 months. One half-day session for a team ended with about 40 scripts against the API, handed over.
  2. Every explanation becomes a hypothesis. Each one gets a test and a verdict. In one 2026 case I tested over 20 hypotheses, one at a time.
  3. The shape before the cause. Google’s own advice is to compare periods and to “click all tabs to find out if the change happened only for specific queries, URLs, countries, devices, or search appearances” (source). I split by URL group and market before I name a cause.
  4. Dates from the data. When the release log has gaps, I date the change from the clicks, URL group by URL group.
  5. Analytics down to the request. When analytics and Search Console disagree, I record the browser’s network requests. In one case I found a consent-tracking bug in those recordings.
  6. A findings workshop, and the scripts. Every number on my slides can be re-run by your team. Ruling a cause out counts as a result, too.

This site, read the same way

In October 2026 I read my own Search Console data before I rebuilt the workshop pages on this site. Property sc-domain:fullstackoptimization.com, the old workshop overview page, 1 July to 18 October 2026:

  • 9,386 impressions, 0 clicks, in 110 days.
  • The average position slipped from about 23 in July to about 36 by mid-October.
  • The query with the most impressions was “seo inhouse”: 101 impressions, average position about 42.
  • “best training for in-house seo teams?” ranked at position 3.6, with 17 impressions and 0 clicks.

My verdict: one overview page ranked for many specific workshop queries, and won none of them. So the in-house training now has its own page, /t/inhouse-seo-training. Whether that pays off, the same data will tell, in two weeks and in two months.

If you want to pull your own data the same way, my step-by-step guide to a Search Console API credentials file is the first step.

Questions I get when traffic drops

Was it the core update?

Maybe. Google itself warns that after an update “there might not be anything fundamentally wrong with your content” (source). I check the update dates against your clicks per URL group. If the drop started before the update, the update is not the cause.

How far back can we look?

Search Console shows 16 months. Google: “If you’d like to extend the 16 months, you could use the Search Analytics API or bulk data exports to pull data and store it in your systems.” (source) Start storing it now, before the next drop.

Is it just seasonal?

Compare the same weeks a year earlier. Google recommends the 16-month view to “make sure it’s not a drop that happens every year due to a festivity or a trend” (source).

Clicks dropped, but impressions did not. What does that mean?

Often a small move in position. Google: “In Search Console, you might see a noticeable drop in traffic without a big change in impressions.” (source) The reverse shape, impressions down and clicks stable, often points to a change in counting or in search features.

What access do you need?

Read access to Search Console, and analytics if you have it. Your release log, however incomplete. Nothing else to start.

How long does it take?

A first verdict on the main hypotheses usually takes days. A full forensic analysis with a findings workshop takes a few weeks.

Sources for the Google quotes on this page, read on 2026-11-12: Debugging drops in Google Search traffic and AI features and your website, Google Search Central, and Data anomalies in Search Console, Search Console Help. The client numbers come from my own audits, workshops and handovers.

From real jobs

  • 2026: two agency analyses tested on every row of Search Console data; the named cause explained about 0 percent of the loss.
  • 2026: a traffic loss tested hypothesis by hypothesis, with changes dated from the data.
  • 2026: AI referrals measured across about 40 sites: between 0.01 and about 1 percent of traffic.
  • 2025: a consent-tracking bug in analytics, found in the browser's network recordings.
  • About 40 scripts against the Search Console API, from one half-day session.