Zero-click search is the pattern where an AI Search engine answers the question on the results page itself, giving no incentive to visit the sourced website. In the first four months of 2026, 68.01% of Google searches ended without a click. It lands hardest on informational, top-of-funnel content, because a definition is exactly what an engine can answer in two lines. Out of that came a convenient theory: informational content no longer pays, so move the content strategy to commercial intent and stop writing glossaries.
We ran this experiment with Conversal, a Belgian digital marketing & AI agency to test if top funnel content still impactful in a zero click world. Their /encyclopedie/ folder is 304 indexable definition pages, a third (33.2%) of everything indexable on conversal.be, measured across four data layers.
Key Findings
Conversal’s got the most cited domain out of 1,400 cited domains in AI Search. 3.0% of all 64,453 citations. Only YouTube came second at 2.7%.
Glossaries got 38x the AI citations vs other pages. 6.03 against 0.16. On Bing, 7.7x.
Bing had x17x higher CTR than Google. The same pages had higher rankings in Bing.
39.8% of all citations go to definition pages. 22.0% folder-only. Copilot leads both.
Only 6.1% of AI crawler visits come from on-demand fetchers. 84.4% are training scrapers.
The industry drew the wrong conclusion from it. SparkToro’s 2026 analysis lists branded, local and transactional queries as the ones still rewarding classical SEO. Informational content is missing from that list, and people read the absence as a verdict.
What this means for your GEO strategy: if your reporting cannot show a citation, you cannot see the return on any top-of-funnel content you publish, and you will keep concluding that it does not work.
Experiment Methodology
A total of 304 glossary pages were made with the following layout to increase AI attractiveness:
The 304 pages under the glossary subfolder /encyclopedie/ were measured across four independent data layers. Nothing here is modelled or extrapolated.
Data Layer
Source
What it captures
Results (16 months)
1. Google Results
Google Search Console
Impressions, clicks, click-through rate and average position in classic Google Search
Which URLs each of the seven AI engines cites, per engine and per URL
AI Citations: 64,453 Domains: 1,400+
4. Agent Analytics
OtterlyAI Agent Analytics, server logs
AI bot visits by bot, engine, URL and category
Crawler bot visits: 13,805
The seven AI Search engines tracked in this GEO experiment are:
Google AI Overviews
Google AI Mode
Gemini
ChatGPT
Microsoft Copilot
Perplexity
Claude
GEO Experiment Results
Layer 1 – Google’s Performance
Google Search Console covers eleven complete months on glossary pages, October 2025 to August 2026. Over that window the 304 glossary pages earned in Google:
Total impressions: 568,000
Total clicks: 565
Average position: 17.9
A click-through rate of 0.10%
Despite many of these pages climbing from an average position of 47.1 to 17.9, a gain of 29 places, and impressions growing close to tenfold, the total number of clicks stayed relatively flat. 565 clicks across the whole period, never more than 101 in a single month. Google gave the folder far more exposure and far better rankings, and almost none of it converted into a visit.
That is the zero-click phenomenon, and definitions are where it bites hardest. A definition is short, factual and has one correct form, so the engine answers above the results instead of sending anyone anywhere..
It is not zero, though. The glossary earned clicks in every one of the eleven months, and the same 304 pages convert at 1.70% on Bing, seventeen times the Google rate, from position 6.0.
Layer 2 – Bing’s Performance
Google vs Bing: Bing has a higher CTR for Glossaries.
The same 304 pages, measured in both search engines on the same impression-weighted estimator.
Seventeen times the click-through rate, and roughly 12 places better. This is the most useful single correction to the zero-click story: the suppression of clicks on definition queries is a Google characteristic, not a property of search. The pattern does exist on Bing, 51 of the 304 glossary pages in its traffic report earned no clicks at all, but it is far weaker there.
Absolute volumes are not comparable, and we are not claiming they are. Google reported 570,367 glossary impressions across 338 days; Bing reported 2,650 in its export window. Bing is a much smaller surface. But it is the surface Copilot cites from, which is why its behaviour on these pages matters out of proportion to its traffic.
Bing’s AI Performance loves Glossary Pages
The glossary pages (/encyclopedie/) is about 33.2% of Conversal’s indexable pages, making it its biggest subfolder on the website. Despite this, it gets in proportion a relatively low number of AI crawler visits compared to other pages. This graph below shows Share of pages vs crawler visits vsBing Webmaster Tools’ AI performance metrics: which is a combination of Microsoft Copilot and Bing’s AI answers only
Blog pages, the homepage, service pages and the robot.txt file and sitemap get crawled a lot more often by AI Agents:
Layer 3 – AI Citations
Conversal’s Glossaries Outrank YouTube and Wikipedia, in Aggregate
A regional agency site beat YouTube, Wikipedia, MDN and Google’s own developer documentation on its own subject matter, with 304 glossary pages that each answer exactly one question.
Conversal also managed to outperform the other websites in brand coverage over time for the specific prompts related to the glossaries.
Copilot Cites the Glossary Six Times More Than Claude
Every engine cites the glossary. What separates them is how much of their citation pool one folder can take.
AI Search Engine
/encyclopedie/ share of citations
domain rank
Microsoft Copilot
6.10%
2
Google AI Overviews
4.74%
2
Google AI Mode
4.50%
2
ChatGPT
3.30%
4
Gemini
2.04%
11
Perplexity
1.43%
4
Claude
1.09%
3
A spread of 5.6 times between top and bottom, on identical content. The reason is source concentration, not format preference. Copilot and the two Google surfaces draw from a small set of domains, so one good folder can take 4.5% to 6.1% of everything they cite. Claude and Perplexity pull from a far wider pool of Dutch-language sources, so the same 304 pages earn a fraction of the share.
Zoom out from Conversal and the format premium is visible across the whole field: roughly 2 in 5 citations in this category (39.8%) go to a definition page on some domain, or 22.0% counting only pages inside a real glossary folder. Copilot leads on both measures. The demand for definitions did not disappear, it moved to a surface most reporting does not cover.
What this means for your GEO strategy: pick the engines where source concentration works in your favour before you pick the tactic. Identical content is worth six times more in Copilot than in Claude.
Layer 4 – AI Crawler Agents
Glossary Pages get 8.8% of total AI Agent visits, but most are AI Training Agents (84.4%).
1209 (8.8%) of a total of 13805 AI Agents crawl the 304 /encyclopedie/ pages.
Of these 1209 AI Agents:
84.4% are AI Training / Data Scraper Bots
9.5% are Search Indexing Bots
6.1% are On-Demand AI Fetchers (Humans asking a question to AI Search)
Nineteen days of crawler logs, 21 August to 8 September, 13,806 AI crawler visits to conversal.be. The glossary took 8.8% of them, against 33.2% of the indexable pages.
AI Search Engine
Share of its own crawl sent to /encyclopedie/
Claude
14.0%
ChatGPT
10.3%
Perplexity
5.4%
Google
2.4%
Microsoft Copilot
0.0%
During the recorded time, Copilot sent 0 of 39 visits to the folder and still cites it more heavily than any other engine. Claude sends the largest share of its crawl there and cites it least. The two behaviours are unrelated.
The category split explains why. Of all 13,806 visits, 84.4% came from training scrapers, 9.5% from search index bots, and only 6.1% from on-demand fetchers, the bots dispatched when a real person asks a question. Of those 839 user-triggered visits, ChatGPT-User is 54.5%, Claude-User 27.7% and Perplexity-User 7.6%
What this means for your GEO strategy: separate the three AI Agent crawler categories before you read anything into log volume. Training scrapers produce no citation, and they are five in six of the traffic. On-demand fetchers are the only visits that represent live demand, and here they are 6.1% which may be the biggest argument why SEO glossaries as a strategy may not be worth the effort.
Conclusion: Should You Invest in Glossary Pages for AI Search?
Short answer: Not for traffic, and not for every company.
Zero click is real and it hit this content type (informational/definition pages) hardest. 570,367 impressions returned 568 clicks, a 0.10% click-through rate, while the pages climbed 29 places in the rankings.
Despite The Zero Click Search Phenomenon, the clicks did not vanish. The same 304 pages convert at 1.70% on Bing, seventeen times the Google rate, from position 6.0. Suppressed clicks on definition queries are a Google characteristic, not a property of search.
Bing’s AI surface rewards them heavily. 33.2% of the indexable site takes 74.4% of Bing’s citations, and Copilot gives the folder 6.10% of everything it cites, more than any other AI Search Engine.
The campaign won its category outright. First of more than 1,400 domains at 3.0% of all 64,453 citations, ahead of YouTube at 2.7% and Wikipedia at 2.3% and all of Conversal’s competitors.
The AI agents that showed up were mostly training scrapers. Of the 1,209 crawler visits the glossary received, roughly 84.4%were AI training and data scrapers, which produce no citation at all. Only about 6.1% were on-demand fetchers, the bots dispatched when a real person actually asks a question.
Worth it for narrow, technical, thinly documented categories where the terminology is searched and competition is thin. Not worth it if the KPI is sessions, or the category already has high-authority reference content.
Follow-up Glossary Video Experiment:
The question is now. Can other formats like video mitigate the Zero Click Phenomenon. As AI Search Engines continuously move in the direction of multimodal answers that include videos in them.
Conversal has agreed to continue the experiment by turning the written entries into video. Our earlier YouTube citation study found video appearing far more often inside Google AI Overviews as clickable results, so it may be the one format that recovers some of the lost click. To be continued.
Got a GEO experiment idea? Let’s collaborate
At OtterlyAI, we don’t write about AI Search from the sidelines. We test it. With the community, we run and update a public GEO Experimentation Tracker showing what we are running and what moved AI Search. If you are running your own GEO experiment, email rick.tousseyn@otterly.ai and we may feature it with your name attributed.
