A 32-day analysis of our Stripe review page GSC data shows query patterns uncommon in human search. 56.6% match a specific syntactic fingerprint.
By Veracity Media Network ยท Published 2026-04-22
From March 20 to April 20, 2026, our TreasuryMetric Stripe review page received 1,390 impressions and zero clicks in Google Search Console. Of the 99 queries that surfaced above GSC's anonymization threshold, 56.6% contained the literal string "2025 2026", 41.4% contained an "evaluat" stem, and 99.83% of counted impressions came from desktop. These patterns do not match the profile of typical human search. They are consistent with automated vendor-evaluation workflows. We cannot prove the source with certainty, but we have ruled out several common alternative explanations. This is a field report, not a definitive study. The Stripe review page is our anchor case; the broader phenomenon likely extends beyond it.
When a vendor like Stripe is searched on Google, the conventional assumption is that a human has typed a query into the search box. For high-profile vendors in 2026, that assumption is increasingly incomplete.
Three distinct cohorts now generate Google-indexed search traffic for vendor pages. Humans still search, of course. Browser-based AI assistants that use Google as their search backend issue queries on behalf of users, with Gemini Deep Research being the most-documented example. Autonomous agents built on developer APIs, including Google Custom Search and other search services, issue queries programmatically as part of multi-step workflows.
The last two cohorts are structurally invisible to website operators. GSC reports impressions and clicks; it does not report whether the underlying query was issued by a human or an automated system. The only way to detect automated traffic is to look for patterns in the query data itself. These are patterns that cluster in ways human search does not.
Stripe is an instructive case for several reasons. It is among the most-researched fintech vendors globally. Its product surface is broad enough that a thorough evaluation must span at least a dozen product lines. And Stripe itself has invested heavily in agentic commerce: the Stripe Agent Toolkit ships with OpenAI, LangChain, and CrewAI integrations, and the company's use case page for agentic commerce explicitly positions Stripe as infrastructure for AI agents making purchases. This is a vendor simultaneously serving AI agents and being researched by them.
What follows is a field report on the second half of that relationship.
Our Google Search Console export covered the 32-day window during which our Stripe review page had measurable search activity: March 20 to April 20, 2026. The page accrued 1,390 total impressions and zero clicks during this period. Ninety-nine queries surfaced above GSC's anonymization threshold, accounting for 596 impressions in the counted set; the remaining 794 impressions belong to queries that were anonymized because of low individual volume.
Three syntactic patterns dominated the counted set.
Pattern 1: The "2025 2026" suffix. 56 of 99 queries (56.6%) contained the literal two-year string "2025 2026" at the end of the query, accounting for 421 of 596 counted impressions (70.6%). A further 4 queries contained "2025 or 2026", for a combined 60.6% of queries. A typical example structure is: [Stripe product] [feature] [pricing] [review] 2025 2026.
Pattern 2: The "evaluate" template. 41 of 99 queries (41.4%) contained the "evaluat" stem, accounting for 180 impressions (30.2%). Six queries matched the specific template evaluate the [industry] company Stripe on [dimension], which is an unusual query construction, more characteristic of a structured prompt fed to an LLM than of a natural human search.
Pattern 3: Product-line-spanning distribution. The 99 queries collectively covered 9 distinct Stripe product lines: Card Issuing (33% of impressions), Payments (15%), Payroll/Contractor Payments (11%), Security and Compliance (9%), Tax (8%), Billing and Invoicing (8%), Payment Links (4%), Marketplace/Connect (3%), and Treasury/Financial Accounts (3%). It is unusual for a single searcher, or even a single organization's procurement team, to research this full breadth of products in a 32-day window. Agentic research workflows that ask "evaluate Stripe across its full product suite" do produce exactly this distribution.
Each pattern independently is suggestive. Together, with 83% of queries containing at least one of the three markers, they are difficult to explain by human search alone.
External evidence that frames the finding. An October 2025 study by Nectiv analyzed 8,500+ ChatGPT prompts and found that the terms most likely to trigger a web search are "reviews", "2025", "free", "features", and "comparison". Commercial-intent prompts triggered searches 53.5% of the time versus 18.7% for informational prompts. The Software sector averaged ~3 fan-out searches per prompt, the highest of any sector they measured. These findings do not prove the source of our queries, but they describe a published LLM-search pattern that is almost identical to what we observed on our Stripe page.
Beyond the query strings themselves, three behavioral signals in our data deviate from typical human search patterns for a B2B SaaS review page.
99.83% desktop share. Of the 596 counted impressions with device data, 595 came from desktop and 1 from mobile. US mobile share of overall internet traffic hovers around 55%, and B2B software research skews more desktop than consumer search but not to this extreme. SparkToro's March 2026 research found mobile accounted for 77% of zero-click searches on Google; our page skews the opposite direction, with desktop generating essentially all impressions.
Zero clicks across 1,390 impressions at avg position 8.7. For B2B pages in positions 4-10, CTR typically falls in the 1-3% range. Our page produced 0%. Even a page experiencing strong Google AI Overview absorption typically retains some click flow. Zero is unusual, particularly for a page with 298 impressions in the United States, where AI Overview absorption and typical human search behavior both predict at least a handful of clicks.
Front-loaded temporal pattern. The first 18 days of activity produced a trickle: 3 to 21 impressions per day. Starting April 8, 2026, volume jumped to a consistent 39-119 impressions per day. This is the kind of step-function pattern that is more consistent with the launch or scaling of an automated research process than with organic human discovery, which tends to build more gradually as a page's position improves.
A responsible field report tests the hypothesis against the strongest alternative explanations. We considered four, and here is how each holds up against the data.
Alternative 1: SEO ranking tools scraping Google with the &num=100 parameter. This is the most plausible-seeming alternative at first glance. SEO consultant Brodie Clark documented that desktop-heavy, zero-click GSC impressions were substantially inflated from early 2025 through September 2025 by third-party SEO tools scraping the full 100-result Google SERP. Growth Hackers' analysis confirmed that when Google disabled the &num=100 parameter on September 12, 2025, impressions across the industry fell 30-70% overnight, with desktop hit hardest. This mechanism produces exactly the desktop-heavy, zero-click pattern we observed.
However, the timing rules this out for our data. Our Stripe review page only began receiving measurable impressions on March 20, 2026, more than six months after the SEO-scraper channel was shut down. The &num=100 bot pattern cannot explain impressions on a page whose entire measurement window begins after the mechanism was disabled. This alternative is well-documented, but it is not our alternative.
Alternative 2: Google AI Overview absorption. When Google's AI Overview renders a synthesized answer at the top of the SERP, a page that is cited receives an impression but often no click, because the user's question is answered inline. This is a real phenomenon that drives zero-click behavior on many pages.
The data is inconsistent with this being the primary cause. Published research from Honcho and Go Fish Digital establishes two facts that weaken this hypothesis for our page. First, AI Overview traffic is ~81% mobile and ~19% desktop, per multiple 2025-2026 studies; our data is 99.83% desktop. If AI Overview absorption were driving the pattern, we would expect a mobile skew, not a desktop skew. Second, B2B software comparison queries (which ours are) are among the least-exposed query categories to AI Overview absorption. Go Fish Digital's March 2026 research found enterprise software and comparison queries experience ~50-60% less AI Overview exposure than general informational queries. AI Overview absorption likely contributes some impressions, but it cannot explain the dominant pattern.
Alternative 3: An audience that is inherently more desktop-heavy than general web users. TreasuryMetric's audience is finance and payments professionals, a population that does lean more desktop than average. This is real.
But it does not get us to 99.83% desktop. Even the most desktop-skewed B2B audiences in published data typically land in the 70-85% desktop range, not 99.83%. A 14-percentage-point excess over the desktop-skewed B2B baseline is unexplained by audience composition alone.
Alternative 4: Stripe-specific search volatility. The page is new, Google is still working out who should see it. New pages often produce unusual ratios in their first few months as Google's algorithms calibrate.
This is partially valid. Some of the pattern we observe is probably attributable to a new page still settling into its rankings. However, it does not explain the specific query syntax (the "2025 2026" suffix, the "evaluate" template), the zero-click rate across 1,390 impressions, or the 9-product-line spread in a 32-day window.
Combined verdict. No single alternative explanation accounts for the full pattern. The SEO-bot hypothesis is ruled out by timing. AI Overview absorption is inconsistent with the device skew. Audience composition explains some of the desktop skew but not its extremity. New-page volatility explains some position jumpiness but not the syntactic fingerprint. What remains is the hypothesis that at least a meaningful portion of this traffic is automated vendor research, whether from AI agents, LLM research tools, or both.
Not every AI agent hits Google. The research literature is specific about which systems use which search backends, and this matters for interpreting what we are seeing.
ChatGPT agents primarily do not use Google. A Search Engine Land analysis of 100 ChatGPT Agent Mode conversations found that the agent used the Bing Search API 92% of the time. Seer Interactive's 500-citation study found 87%+ of SearchGPT citations matched Bing's top organic results. ChatGPT agent traffic therefore does not flow through GSC at all; it shows up in Bing Webmaster Tools. For our purposes, ChatGPT agents are not the likely source.
Gemini Deep Research uses Google explicitly. Google's own Gemini API documentation documents that Gemini Deep Research agents use the google_search tool and issue approximately 80 search queries for a typical research task, up to 160 for complex competitive analyses. Every one of those queries is a Google query and will appear in GSC for any page ranked for those terms. This is the most well-documented Google-issued agentic search channel in the market today.
Custom agents via MCP or developer APIs. Google Custom Search API offers a free tier of 100 queries per day and $5 per 1,000 queries beyond that. Any developer building a research agent can point it at Google for a few dollars per 1,000 queries. These agents would show up in GSC with the querying patterns their creators encoded in prompts.
Perplexity's crawlers are a separate mechanism. Cloudflare's August 2025 research documented that Perplexity uses undeclared "stealth" web crawlers to bypass web application firewalls, a different channel than Google search. Perplexity would generate crawler traffic, not Google SERP impressions, unless it uses Google as a search engine under specific flows.
Given this landscape, our hypothesis narrows. The most plausible producers of the pattern we observed are Gemini Deep Research, custom agents using Google Custom Search or scraper-based approaches, and residual human search (which is always present but cannot account for 56.6% of queries bearing a specific two-year-string suffix).
Three practical implications follow from a dataset like ours, regardless of whether our interpretation holds.
Detecting automated search in your own GSC. The heuristic is simple: export your per-query data from any high-profile review page and check what percentage of queries contain (a) explicit year-range suffixes like "2024 2025" or "2025 2026", (b) the words "evaluate" or "review", combined with "features" or "pricing", and (c) a product-line-spanning distribution. If these patterns dominate, and your clicks are low relative to impressions, and your device mix is heavily desktop on a B2B page, some portion of your impressions is probably automated. No specific threshold separates "human" from "automated" cleanly, but the higher the percentage of queries with these markers, the more of your traffic is likely not human.
Whether to optimize for it. Yes, but cautiously. Automated research that doesn't click is still valuable: the LLM synthesizing the answer may cite your page, and that citation reaches a human decision-maker downstream. The best way to be cited is to publish specific, structured, verifiable data that an LLM can extract: concrete numbers, tables, methodologies, update dates, and original calculations. This is the advice currently given for LLM citation optimization, and it aligns with our own pattern observations. The piece you are reading is structured accordingly.
Whether to try to filter it out. Probably not. Automated queries and human queries are both counted as impressions by GSC and both influence Google's understanding of your page's relevance for those terms. Filtering out automated traffic, even if possible, would suppress signals that help your page rank for related human queries.
Stripe is unusual in the vendor landscape because it is both the subject of agentic research queries (per the data above) AND the payment infrastructure for agentic commerce. Both halves of this are relevant to how the broader phenomenon will evolve.
On the infrastructure side: Stripe launched the Agent Toolkit, an open-source SDK with integrations for OpenAI, LangChain, and CrewAI, in mid-2024. The toolkit lets agents create payment links, issue invoices, and charge cards programmatically. Stripe has since built this out into the Agentic Commerce use case, explicitly positioning the company as the payments layer for AI-initiated transactions. Adyen's agentic commerce positioning follows a similar but slightly different path, emphasizing protocol-level readiness (including the Agentic Commerce Protocol) over shipping a specific toolkit.
On the research side, which is what our field report is about, Stripe is being evaluated by research agents at measurable volume. We have 596 counted impressions across 9 product lines in 32 days on a single review page, with a query profile overwhelmingly consistent with automated vendor research.
The two phenomena reinforce each other. As more businesses deploy agents that buy things on Stripe, more vendor-evaluation queries get issued to Google by agents researching which payment infrastructure to use. As Stripe's share of agent-initiated payment volume grows, so does the research volume on its product surface. This is not speculation on our part; both sides of the loop are documented. The only open question is how fast the loop accelerates, and our 32-day snapshot only catches the very beginning of it.
For reference, here is the full query classification from the dataset. Impression aggregates and percentages are safe to publish; individual query strings are not.
| Stripe product line | Share of impressions | Average position | Share of 99 queries |
|---|---|---|---|
| Card Issuing | 33% | 8.4 | 32% |
| Payments | 15% | 8.6 | 16% |
| Payroll / Contractor Payments | 11% | 12.9 | 11% |
| Security / Compliance | 9% | 6.0 | 8% |
| Tax | 8% | 7.4 | 9% |
| Billing / Invoicing / Subscriptions | 8% | 11.0 | 10% |
| Payment Links | 4% | 8.2 | 5% |
| Marketplace / Connect | 3% | 17.7 | 4% |
| Treasury / Financial Accounts | 3% | 5.4 | 3% |
Syntactic pattern distribution, share of 99 queries (596 counted impressions):
- "2025 2026" literal suffix: 56.6% of queries, 70.6% of impressions
- "2025 or 2026" variant: 4.0% of queries, 4.7% of impressions
- "evaluat" stem anywhere: 41.4% of queries, 30.2% of impressions
- "features" AND "pricing" in same query: 25.3% of queries, 38.3% of impressions
- "evaluate the [industry] company Stripe" template: 6.1% of queries, 3.9% of impressions
- Queries without any of these markers: 17.2% of queries, 10.2% of impressions
Position distribution (impression-weighted): 5.0% in positions 1-3, 65.1% in positions 4-10, 28.5% in positions 11-20, 1.3% in positions 21+.
Device and geography: Desktop 99.83% / Mobile 0.17% of counted impressions. Top 3 countries by impression: United States (50%), Brazil (5.9%), Canada (4.2%). Zero clicks across all countries and devices.
We will refresh this analysis quarterly. Next update: 2026-07-22.
Data source: Google Search Console performance export for https://treasurymetric.com/review/stripe/, filter "Last 3 months", exported 2026-04-22. Active measurement window within that filter: 2026-03-20 to 2026-04-20 (32 days). Impressions: 1,390 total page-level, of which 596 are attributable to specific queries above GSC's anonymization threshold. The remaining 794 impressions (57.1%) were aggregated as anonymized long-tail queries.
Classification process: We regex-matched each of the 99 surfaced queries against the pattern definitions described above. Every reported percentage in this piece is a direct count from the xlsx export. We did not infer any data; we counted it.
What we verified externally: The Nectiv (October 2025), Search Engine Land (October 2025), Seer Interactive (February 2025), Cloudflare Radar (August 2025), Brodie Clark (September 2025), Honcho (March 2026), and Go Fish Digital (March 2026) studies referenced in this piece were all reviewed before drafting. Each citation points to published research with a specific methodology and date.
What we did NOT verify: We cannot distinguish queries issued by humans from queries issued by automated systems using GSC alone. No GSC field records user-agent, referrer, or automated/human distinction for search queries. The attribution of queries to specific AI systems in the "Which AI Systems" section is directional, based on each system's documented behavior; we did not trap or log any specific automated session.
Affiliate disclosure: Veracity Media Network has no affiliate partnership with Stripe, Adyen, or any payment vendor referenced in this field report, as of 2026-04-22.
Update cadence: Quarterly or on any material finding that updates the hypothesis.
TreasuryMetric is part of the Veracity Media Network, six independent SaaS auditing directories covering AI, cybersecurity, finance, HR, logistics, and energy software. The network audits 800+ tools using the Compliance Transparency Index, a proprietary 0-100 score evaluating security posture, pricing transparency, data rights, and corporate legitimacy. TreasuryMetric currently audits 71 finance and payments tools.
Our philosophy is "Documentation, Not Opinions." This field report reflects that: we report what we observed in our data, what external research establishes about AI search behavior, and which hypotheses our data supports versus rules out. We do not claim a definitive conclusion. We commit to testing the hypothesis again in 90 days.
If you operate a vendor review site or directory and want to compare notes on search patterns in your own GSC data, or represent Stripe or any vendor discussed and want to correct information, reach out via our vendor contact process on TreasuryMetric.com.