Double Atari SEO progress report: what 90 days of data actually says

What do the numbers actually say?

All figures below come from Google Search Console, GA4, and Semrush for this site, covering June 18 to September 15, 2026, against the preceding 90 days.

Metric	Last 90 days	Prior 90 days	Change
Impressions	2,007	292	Up 587%
Clicks	10	14	Down 29%
Average CTR	0.5%	4.79%	Down
Average position	25.0	10.1	Worse
Authority Score	2	2	Flat
Referring domains	220	220	Flat
Organic keywords	3	3	Flat

If a client handed me that table with no context, I would say one metric moved and the rest are consequences of it.

Why did impressions grow 587 percent while clicks fell?

Because I published a lot, those pages got indexed, and they landed at positions deep enough to generate impressions without ever earning a click. This is the most misread pattern in Search Console, and it looks like failure while being mostly mechanical.

The page-level data makes it obvious:

• /blog/wpvibe-ai-vibe-coding-wordpress-websites/: 428 impressions, 1 click, average position 11.2
• /services/seo-consulting/: 340 impressions, 0 clicks, average position 51.2
• /blog/geo-vs-aeo-vs-seo/: 230 impressions, 0 clicks, average position 41.7
• /minneapolis-seo/: 204 impressions, 0 clicks, average position 26.0
• Homepage: 335 impressions, 8 clicks, average position 4.3

Position 51 produces impressions and no clicks, by definition. Add several pages like that to a small site and the site average position collapses while the site CTR is divided by a much larger impression base. Nothing got worse. The denominator changed.

The homepage is the control that proves it. Still at position 4.3, still carrying 8 of the 10 clicks. Where I actually rank, the clicks still come.

The genuinely useful finding in that table is a different one: /about drew 309 impressions at average position 3.7 and zero clicks. That is not a depth problem, that is ranking well and not being clicked, which is the one place on this site where a title and description test is clearly warranted. It is the cheapest available win here, and it is the only CTR test I plan to run this quarter.

Why is the Authority Score stuck at 2?

Here is where I have to correct something I believed going into this analysis, and it is the most useful part of this post.

Semrush calculates Authority Score from three components: Link Power, which is the quantity and quality of referring domains, Organic Traffic, and Spam Factors. Link Power carries the most weight, followed by organic traffic.

My profile: 220 referring domains, 727 backlinks, 509 follow and 218 nofollow, spread across 223 IP addresses but only 147 class C subnets. Three organic keywords. Effectively zero estimated organic traffic.

So the score is 2 because two of the three inputs are close to empty. I have almost no organic traffic, and while 220 referring domains sounds respectable, the overwhelming majority are worthless. Only three links in the entire profile are ones I would defend: my own portfolio site, another domain I own, and a paid newsletter sponsorship.

That is the real diagnosis. The constraint is the absence of good links, not the presence of bad ones. I had those two things backwards, and the disavow file is what forced me to notice.

The disavow file I built and then decided not to upload

I ran a full audit of all 220 referring domains and classified every one. The result was 211 domains to disavow, 6 held for review, and 3 to keep. The spam was not subtle:

• A 48 domain private blog network on a single Singaporean IP range
• A 16 domain link shortener farm sharing one Moldovan IP address
• 13 parked domains on sequential GoDaddy IPs
• 5 paid link vendors sharing one IP
• One expired-domain blog throwing 84 sitewide links at me

I verified the borderline cases by fetching them. One was a podcast domain that had expired and become a Vietnamese gambling site. One described itself as a private blog network in its own copy. So the file was accurate and the work was sound.

Then I checked whether uploading it would do anything, and the answer was no. Two reasons.

A disavow file cannot change your Authority Score. Disavowing is an instruction to Google. Semrush never sees it. As one Google Search Central response put it directly, disavow "has zero influence on 3rd party tools or their metrics." So if your goal is a higher Authority Score, the disavow tool is the wrong instrument entirely. I had been treating a Google-facing tool as a fix for a Semrush-facing number.

Google says most sites should not use it. The official documentation is specific. You should disavow only if both conditions hold: you have a considerable number of spammy, artificial, or low quality links, and those links have caused a manual action, or likely will. It also says plainly that "in most cases, Google can assess which links to trust without additional guidance, so most sites will not need to use this tool," and warns that used incorrectly the feature "can potentially harm your site's performance in Google Search results."

I meet the first condition comfortably. I have no manual action, so I do not meet the second. And there is a practical blocker on top of that: the disavow tool requires property ownership, and my Search Console access on this property is full user rather than owner, so I would need to verify ownership before I could upload anything.

So the file sits on disk, finished and unused. That is not wasted work. It is documentation of exactly what my link profile is, ready to go if a manual action ever appears.

So what should you actually do about spam links?

The sequence I would now give any client, in order:

1. Check Search Console under Security and Manual Actions first. If it says no issues detected, spam links are not your problem and you should stop investigating them. This takes thirty seconds and it reframes everything downstream.
2. If there is no manual action, do nothing about the spam. Google discounts this kind of link automatically. Negative SEO through link spam is far rarer than the tools selling toxicity scores imply, and a toxicity score is a vendor's opinion, not a Google signal.
3. Audit and document anyway. Know what is pointing at you. Keep the file current. You want it ready before you need it, not built in a panic.
4. Put the effort into earning links instead. This is the actual constraint and it is the harder, slower, correct answer.
5. Only upload a disavow file if you have a manual action, or you are genuinely about to, for instance because you bought links in the past and know it.

If you take one thing from this post: a disavow file is a remedy for a penalty, not a growth tactic.

The opportunity I found while looking for the actual fix

If the constraint is earning links, the question becomes what would earn one. So I went looking for a topic with real demand and weak competition where I could publish something genuinely better, and the keyword data was unusually clear.

Keyword	US monthly volume	Competition
technical seo audit	5,400	0.03
technical seo audit service	3,600	0.01
seo technical audit	2,400	0.03
technical seo audit checklist	2,400	0.01
seo technical audit checklist	1,900	0.01
technical seo audit template	480	0.01

That is over 16,000 monthly searches in one tight cluster at almost no advertiser competition. More importantly, checklists earn links, because people reference them.

The competitive picture said the same thing. Moz's audit checklist is comprehensive at roughly 70 items but carries a 2021 date and does not address AI search at all. Semrush has updated theirs to include AI grounding, but it runs to about 23 items. Sitebulb gates most of its content behind a template download.

So the gap was not item count. It was sequencing and integration. Every checklist I looked at is a flat list sorted by category, which quietly invites teams to work top to bottom and spend three weeks compressing images while a canonical conflict keeps half the site out of the index. So I wrote the checklist in dependency order, in three tiers, with AI search items integrated throughout rather than bolted on as a final section, and a verification step attached to every item.

One strategic note I would give any client looking at that keyword table: the "service" and "services" variants carry a $13.79 CPC, which is commercial intent, and they should point at a service page rather than a blog post. Do not aim a checklist article at a buying query.

Why LLM SEO and AI visibility tracking are next

The same gap analysis surfaced a second cluster, and this one sits directly on what I already do.

Keyword	US monthly volume	CPC
ai search optimization	3,600	$17.70
llm seo	1,600	$10.54
chatgpt seo	1,600	$7.65
aeo agency	1,000	$21.16
llm visibility	1,000	$10.50
ai visibility tracking	720	$12.42
ai seo consultant	210	$18.99

Two things stand out. The CPCs are high, between $10 and $21, which is the market telling you these searches convert. And in Semrush's 12 month trend data, both llm visibility and ai visibility tracking are at their peak right now, as is aeo agency. These are not mature terms, they are terms still on the way up.

So I wrote two posts rather than one, because the cluster contains two genuinely different intents.

LLM SEO: how language models decide what to cite covers the mechanism. Almost everything published on this topic is a tactics list with no explanation of the retrieval pipeline those tactics are meant to influence, which is why so much of it is wrong. Writing the mechanism down also let me address two popular tactics that do not survive the evidence, including llms.txt, which the major AI crawlers largely do not request.

AI visibility tracking covers measurement, and it is the more urgent of the two because the ground moved this quarter. Google shipped generative AI performance reports in Search Console in June 2026 and completed the worldwide rollout on August 31, then acknowledged in September that the reporting is inadequate. That is a live, specific, useful thing most site owners have not looked at yet. It also plays to the part of this work I am strongest at, which is measurement rather than opinion.

The deliberate choice in both: I already have posts on ranking in ChatGPT and Perplexity, AEO, and GEO. Writing another general "how to show up in AI search" piece would have competed with my own pages. Mechanism and measurement were the two angles I did not already own.

I am not writing anything for aeo agency or ai seo consultant despite the $21 and $19 CPCs, because those are buying queries. They belong on service pages.

What does GA4 say about the people who do arrive?

Split the way I split it for clients, All Traffic and US Only, with engaged sessions included:

Channel	All sessions	Engaged	US sessions	US engaged
Direct	258	83	190	66
Organic Search	131	66	130	65
Unassigned	31	18	31	18
Referral	30	10	21	10
AI Assistant	2	1	0	0
Organic Social	1	0	0	0
Total	453	178	372	159

Organic Search converts to engagement at 50 percent against Direct's 32 percent. Small numbers, but the ordering is the point: the fewer visitors who arrive from search are considerably more interested than the larger direct group, which supports putting effort into search rather than into broadcasting.

Two sessions in AI Assistant is worth noting precisely because it is so small. That channel exists now, it is real, and anyone judging AI search investment by referral sessions will conclude it does not work. That is the wrong read, and it is why the measurement layer matters.

What happens over the next 90 days

Concrete, in order:

1. Check Manual Actions, then leave the disavow file alone. Verify property ownership so the option exists, and do not use it.
2. Test the /about title and description. Position 3.7, 309 impressions, zero clicks. One page, one change, measurable. I tried a broad sitewide CTR rewrite earlier this year and reverted it, because changing everything at once means learning nothing.
3. Publish the three posts in this cycle and watch the generative AI report in Search Console for whether they get picked up in AI features.
4. Fix the internal linking to the service pages. /services/seo-consulting/ sitting at position 51 with 340 impressions is a page search engines can see and have not been given a reason to trust.
5. Actually pursue links. The checklist is the asset. Publishing it is not a link strategy, it is the thing a link strategy needs.
6. Filter proxy and preview hostnames out of GA4 so the numbers above stay trustworthy as the base grows.

The single sentence version: stop optimizing a site nobody links to and start earning the links, while fixing the handful of pages that already rank and are not being clicked.

Progress report FAQ

Why did impressions go up while clicks went down?

New pages got indexed at positions deep enough to generate impressions without earning clicks. Adding pages at positions 26 to 51 to a small site raises the impression base sharply, which drags down both average position and site-wide CTR without anything actually getting worse. The homepage held position 4.3 and still produced 8 of the 10 clicks.

Will a disavow file improve my Semrush Authority Score?

No. Disavowing is an instruction to Google, and Semrush never sees it, so it cannot change a Semrush metric. Google Search Central has stated that disavow has zero influence on third-party tools or their metrics. If your Authority Score is low, the cause is usually weak link power and low organic traffic rather than the presence of spam links.

When should you actually upload a disavow file?

Google says to disavow only if both conditions apply: you have a considerable number of spammy, artificial, or low quality links, and those links have caused a manual action or likely will. Check Security and Manual Actions in Search Console first. If it reports no issues detected, you almost certainly should not upload one.

How do you raise a low Authority Score?

Earn quality referring domains and grow organic traffic, because those are the two heaviest inputs in the calculation. Removing or disavowing spam links does not substitute for that. In my own case, 220 referring domains produced a score of 2 because only three of them were links worth having.

Why write a technical SEO audit checklist instead of something else?

Because the keyword cluster carries over 16,000 US monthly searches at competition between 0.01 and 0.03, checklists tend to earn links, and the existing pages had a clear gap. The strongest competing checklists are either flat category lists with no AI search coverage or short summaries, so sequencing the checks in dependency order was a genuine differentiator.

Why publish your own site's weak numbers?

Because the diagnosis is the product. Anyone can show a chart going up. Showing a real profile, explaining why a plausible fix was the wrong one, and documenting the correction is a more honest demonstration of how the work actually gets done.