Open Google Search Console right now and there is a decent chance you are looking at a chart that makes no intuitive sense. Impressions climbing. Clicks flat or falling. The gap between the two lines widening every week.
Parth, who co-hosted Episode 134 alongside Malhar, has a name for it borrowed from the SEO community: the crocodile mouth. Impressions keep rising. Clicks stay flat. One jaw moves up, the other barely moves, and the gap between them just keeps growing.
This episode was about diagnosing that gap, separating the causes that are within a practitioner’s control from those that are structural shifts in how Google works, and figuring out what to actually do about it.
GSC Impressions and Traffic Difference – Not A New Problem
The first instinct when you see impressions rise and clicks fall is to assume something broke. But Parth’s opening point was important: this problem predates AI Overviews by several years. It started with featured snippets, where Google began answering queries at the top of the page without requiring a click. Then came Search Generative Experience. Now AI Overviews. Each layer added more on-SERP answers, more reasons for a user to stay on Google rather than visit a website.
What changed recently is not the mechanism but the scale. The traffic numbers that used to absorb this quietly, because there were enough clicks left to sustain a baseline, are no longer sufficient to hide what is happening. The numbers are now visibly, undeniably off. That is why the crocodile mouth has become a talking point. Not because the jaw opened recently, but because impressions have climbed far enough above the click line that the gap is impossible to rationalize away. And as we explored in an earlier session, most SEO teams are too quick to blame AI Overviews for falling traffic when the real causes sit much closer to home.
There is a second, newer cause layered on top: query fan-out. When someone types a query into Google’s AI mode, the system internally fires multiple related sub-queries to gather context before generating an answer. Each of those sub-queries can generate impressions for your pages. Your impressions go up. But the underlying search was never a human user who intended to click through to your site. It was a machine gathering signal.
Add log file analysis to the picture and another layer becomes visible: LLM bots from ChatGPT, Perplexity, and other AI search platforms are crawling websites at increasing rates. Those crawl requests can, in certain configurations, surface as impressions data. The number on your GSC dashboard is no longer a clean count of human searches that showed your page. It is a composite of human queries, AI fan-out sub-queries, and bot crawl activity. Reading it the way you read it in 2022 will lead you to the wrong conclusions.
Position One Is No Longer the Finish Line
One of the sharper questions Malhar asked: if you are ranking number one and still losing clicks, does that position still mean anything?
Parth’s answer: it is still important, but it is no longer the end goal it used to be. Ranking at position one or position zero through a featured snippet was, until recently, the highest value outcome an SEO could deliver. It meant maximum CTR. In the current SERP environment, ranking at number one for an informational query mostly means you will appear in the AI Overview citation set. You might get mentioned. You almost certainly lose the click.
The more relevant question now is not where you rank but whether you are part of the answer at all. Are you cited in the AI Overview? Do you show up in the “people also ask” expansion? Are you referenced in the YouTube videos that surface when someone searches? These are not the same as position one, but they are increasingly where the influence happens.
Vijay, who shared live data from his own research during the session, confirmed this with a concrete finding: YouTube is the highest cited domain in AI Overview results. His experiments across both SEO-related queries and e-commerce queries consistently showed YouTube video content being pulled into AI answers at a higher rate than any other domain type. And here is the detail that sharpens that finding: the majority of those cited YouTube videos were not from brand-owned channels. They were from third-party creators who had built topical authority in the category. For most brands, their own YouTube channel was cited in only 3 to 7 percent of relevant AI Overview responses. The rest went to independent creators.
The implication is significant. Creator partnerships and influencer outreach are not soft brand activities anymore. They are an AI citation strategy.
Are You Ranking for the Wrong Things?
The crocodile mouth problem has two distinct causes that look identical in GSC. One is a SERP layout shift: your rankings are stable but AI Overviews, ads, and other features have pushed your links below the fold or absorbed your clicks. The other is a relevance problem: you are ranking for topics that were never closely connected to your business, attracting impressions from an audience that was never going to convert.
Both show up as impressions-up, traffic-down. But they require completely different responses.
Parth’s diagnostic approach starts with filtering GSC data for pages that get over 5,000 impressions per month but fewer than 10 clicks. A page in that state is either a victim of SERP layout changes or it has drifted into irrelevant territory. The next question is whether competitors in the same category are experiencing similar traffic trends. If your traffic is falling while competitor traffic holds steady for similar keywords, the problem is specific to your site or content. If the whole category is declining, it is a structural SERP shift and everyone is in the same position.
The Backlinko example Parth raised is instructive. Before its acquisition, Backlinko had roughly 170 to 180 published blog posts. Almost every single one was generating meaningful traffic. The density of organic traffic per published page was high and maintained across the site. That ratio, traffic distributed across most pages rather than concentrated in a handful, is a signal of genuine topical relevance. When a large portion of your published content generates impressions but no clicks and appears in no AI surface citations either, that is the content audit conversation from Episode 132 resurfacing: you have accumulated pages that were never closely enough tied to your actual business or audience.
The response options are two: rework the content to better match the current format reality (embed a YouTube video, tighten the intent match, strengthen the link between the page and your core offering), or prune. Not every content piece deserves a rescue. If it cannot be connected to the customer journey or to your core product, remove it.
Should You Celebrate Visibility Without Traffic?
Malhar asked this deliberately. It is a slightly uncomfortable question because the honest answer is yes, but with conditions most people skip past.
Parth’s framing was measured.
Losing direct traffic while maintaining visibility is a loss relative to what organic used to deliver, and that is worth acknowledging honestly. Google is training on publisher content and answering queries with that content, while sending fewer visits back to the publishers who created it. That is a real tension and the publishing community has been vocal about it.
As we covered in depth in Ep. 133, SEO is no longer a traffic channel, it is a brand influence channel and that shift changes how success needs to be measured entirely.
At the same time, if everyone in a category is in the same position, the competitive question is not “am I losing traffic” but “am I losing ground to competitors.” Are you still showing up across the AI answer surfaces where your target audience is researching? Are your brand mentions in AI Overview answers proportionally higher or lower than your competitors? Share of voice across AI surfaces is becoming the metric that matters, even before a click occurs.
The caveat: impressions that come from irrelevant queries or query fan-out sub-queries are noise, not signal. Impressions from genuine branded or category queries where your brand appears and makes an impression on a real user are meaningful even without a click. The two need to be separated before deciding what the impressions number is actually telling you.
Is Google Over-Reporting Impressions?
Parth’s answer here was direct, and worth quoting closely: yes, the impression numbers are skewed.
The combination of query fan-out, LLM bot crawls, and multi-surface indexing means that the impression count in GSC is not a clean representation of human eyes on a search result. Google is in a transition period where it has not yet built out full AI search dashboards for Webmaster tools globally. The AI Search Console features that show citation and AI Overview data are still rolling out, primarily in the US and a handful of Western markets. Until that data is available to practitioners worldwide, impressions remain the primary metric on the dashboard, and they are increasingly noisy.
Parth’s description of this was blunt: Google is occupying SEO practitioners with another vanity metric while the more meaningful dashboard data catches up.
Vijay pointed to Bing Webmaster Tools as a meaningful contrast. The roadmap shared by Bing’s AI search lead, Krishna Madhavan, at the SEO Week conference showed measurement infrastructure that is meaningfully ahead of what Google is currently offering Webmasters. The cancelled Google I/O meeting between veteran SEOs and the Google Search team was a frustrating data point in the same direction.
Is GSC Data Enough to Diagnose the Problem?
Not on its own. Parth’s estimate was that GSC gets you about 80 percent of the diagnostic signal you need. But the remaining 20 percent requires connecting it to other sources.
The SERP layout dimension specifically cannot be properly diagnosed from GSC alone. You need to manually check how the SERP looks for your key queries, track how that layout has changed over time, and note where AI Overviews, ads, and other features are physically displacing your result. CTR data in GSC is increasingly unreliable as a standalone signal precisely because impressions are inflated by non-human activity. A falling CTR may mean nothing changed about your result except that Google put more things above it.
For brands running paid campaigns, cross-referencing Google Ads keyword performance data with GSC query data reveals which segments are absorbing the clicks that used to go to organic. That combined view gives a more accurate picture of what users are actually clicking when they see your brand across the SERP.
Parth’s practical recommendation for anyone willing to go deeper: set up a GSC bulk data export to BigQuery and start storing query-level data beyond the 16-month rolling window that the GSC dashboard allows. GSC only shows you data from when you set up the export forward, but you can backfill historical data using the GSC API, combine it with the ongoing export, and build a dataset that supports much more sophisticated analysis, including identifying query-level trends over multiple years rather than months. The analysis itself is significantly easier now, he added, with tools like Claude Code that can work through large SQL query outputs without requiring deep manual query-writing skills.
The Branded Traffic Anomaly
A community member Suganthan, brought a specific problem to the panel: branded search traffic declining in GSC, with no paid ads running on branded terms. What causes this and how do you diagnose it?
Parth walked through the diagnostic checklist. External PR campaigns or creator partnerships from a few months ago can spike branded search temporarily and create a comparison baseline that looks like a decline when the campaign effect fades. Spelling variants or name collisions, where your brand name resembles another term being searched in a different context, can distort branded impression and click data. Site migrations, changed URL structures, or shifted support documentation can redistribute branded clicks across different pages in ways that look like drops without being actual declines.
The broader point: branded search demand is not only an SEO outcome. It is a downstream effect of everything else the brand is doing: offline presence, influencer work, product launches, media coverage. Tracking it properly requires a 16-month minimum view to separate genuine trend from noise, and cross-referencing it against any external brand activity from the same period.
What Should You Do This Week?
Malhar closed the session the way he always does: if there is one concrete thing practitioners should do after this conversation, what is it?
Parth’s answer: set up the GSC bulk data export to BigQuery, today if possible. The UI dashboard gives you a filtered, simplified view of a much richer underlying dataset. The bulk export gives you access to query-level data at a granularity that the dashboard intentionally hides. Combining it with AI tools to navigate the volume of data means that this level of analysis is no longer limited to teams with dedicated data engineers. It is accessible to any practitioner willing to set it up.
For everyone who is not yet at that level of tooling, the minimum diagnostic before drawing any conclusions from a crocodile-mouth chart is to separate impression growth by source: branded versus non-branded, informational versus commercial, known category queries versus queries where your pages are clearly appearing for off-topic reasons. The problem you are solving for looks entirely different depending on which category is driving the divergence.
The Bigger Picture
This episode, more than most, was a diagnostic session rather than a strategy one. The honest takeaway is that the metrics SEOs have relied on for a decade are becoming harder to interpret in isolation. Impressions mean multiple things now. CTR is less reliable as a performance signal. Position one is a starting point, not a destination.
What has not changed is the underlying logic: be present wherever your audience is making decisions, earn the trust of the platforms and communities that feed into AI surfaces, and make sure the content you have published is genuinely tied to what your business does and who it serves.
The crocodile mouth is real. Impressions will keep rising. But clicks will not follow unless the content earning those impressions is genuinely tied to what your audience is trying to decide.
SEOTalk Spaces is a weekly community conversation hosted by Malhar Barai and Parth Suba. Episode 134 featured contributions from Vijay and community members from the SEOTalk audience.
Check out the recap: