The 20-Line Test: If Google Can Generate Your Site, It Will
TL;DR: On September 7, an interactive character counter appeared inside Google's AI Overviews. Not a link, not a snippet — a working tool, maybe twenty lines of JavaScript, executing inside the search results. Every website whose business is that tool — sites ranking #1 for queries worth millions of monthly searches, monetized with ads — just lost its reason to exist. This is not zero-click search. It is zero-need search, and it collapses the moat of every utility site built on SEO. Here's the test that tells you, before Google does, which opportunities are already dead: if a model can generate your core functionality in twenty lines, you don't own a product. You own a feature Google hasn't shipped yet.
James here, CEO of Mercury Technology Solutions.
Hong Kong — September 2026
The Twenty Lines That Ended a Business Model
Imagine building a website, ranking #1, collecting millions of searches, monetizing with ads. Then Google writes roughly twenty lines of functionality — and your reason to exist disappears.
That's not a hypothetical anymore. Search-industry coverage on September 7 reported an interactive Character Counter widget loading directly inside AI Overviews for character-counter queries. The user types "character counter," the tool appears, the counting happens, the user leaves. No click. Not because the user chose not to click — because the click no longer has a purpose.
Understand what that archetype site actually was. A character counter is deterministic, stateless, and universally needed: the exact query profile that accumulates millions of searches a month. AdSense on top of that volume built small empires. The SEO was immaculate — backlinks, page speed, topical authority. It ranked #1 for a decade.
And none of it mattered, because the moat was never a moat. It was a rental from Google, and the landlord just moved in.
Not Zero-Click. Zero-Need.
The industry keeps calling this zero-click search. The vocabulary is a decade stale. Zero-click used to mean Google answered the question — capital of France, Paris, done. Informational. The character counter is different in kind: the user wasn't looking for information. They wanted to do something. Google didn't replace the answer. It replaced the interaction.
AI Overviews stopped being an answer layer in 2026. They are becoming an action layer. The machinery is already public: Canvas in AI Mode, rolled out to every US user in March with Gemini 3 underneath, generates working applications — calculators, converters, quizzes, visualizations — directly in the SERP, in seconds, refined through conversation. Google's own product logic is airtight: if the task is small, deterministic, and safely executable in the results page, sending the user elsewhere is a worse experience. They're right. That's what makes it fatal.
The old question — "how do we become the best result for this query?" — is no longer the question. The new question: does Google still need your page to satisfy the query at all?
When the skin is gone, where can the hair attach? Ten character-counter sites competing on domain authority were fighting over rankings. The moment the function moved into the SERP, the rankings became the hair on a skin that no longer exists.
The fortifications were Maginot Lines: perfect against the last war, invisible to the new doctrine. Not one of those sites was outcompeted. They were bypassed.
The 20-Line Test
Strip out the noise and you get a brutal, portable test for any SEO opportunity in the AI era. Before you invest a quarter of content and engineering into ranking for anything, ask:
The 20-Line Test: Describe your page's core value in one sentence. If a model can implement that sentence as a working widget — twenty lines, no APIs, no data, no state — Google will eventually ship it inside the SERP, and your traffic will be the exhaust of that decision.
Character counters. Word counters. Case converters. Unit converters. Password generators. Timer sites. Percentage calculators. Age calculators. BMI calculators. Color pickers. QR generators. Lorem ipsum. Every one of them fails the test. Industry analyses were already projecting 40–60% CTR declines for tool and utility sites from Canvas-class features — before the widgets started arriving inside AI Overviews themselves.
The test scales up, too. Basic template pages — resume templates, invoice templates — describe in one sentence, generate in one prompt. Simple free-tier SaaS tools: same sentence, same prompt. If your free tool exists to capture SEO traffic and could be rebuilt by the model in an afternoon, it wasn't a moat. It was a countdown.
The Moat Map: Where Canvas Can't Follow
Here's the part the panic coverage misses. Google published its own list of what it can't absorb, disguised as Canvas's technical limitations:
- No external API calls — nothing requiring live third-party data
- No persistent storage — nothing with accounts, history, settings
- ~50KB code ceiling — nothing with large datasets or complex logic
- No guaranteed real-time data — the tax calculator ships with last year's brackets
- Single-page only — nothing with multi-step workflows
Read that list again. It is a map of every place a website can still be necessary. Every Canvas limitation is a moat specification. Real-time market data. Proprietary datasets. Authentication and persistence. Multi-step workflows connected to systems of record. This is where durable search visibility lives now — not in ranking for tasks Google can execute, but in being structurally impossible to inline.
The Citation Rule
For the GEO and GAIO work we run at Mercury, one rule now governs strategy:
The Citation Rule: an answer engine will either cite you or replace you — and which one it does is decided by what your page *is*, not how well it ranks.
Answer engines cite sources for claims, data, and expertise. They do not cite character counts. A query whose SERP result is self-contained — a tool output, a definition, a conversion — is a zero-citation query, and zero-citation queries are where utility sites go to die. The escape is not better SEO on the same asset. The escape is changing what the asset is: from utility to evidence. Answer Assets backed by proprietary data — your cohorts, your telemetry, your named experts, your original frameworks — are the only class of content an answer engine must reference rather than replicate.
Stop building sites a model can generate. Start building data a model must cite.
Run the Audit This Week
Four moves, in order:
- Classify your query portfolio. Tool queries (execute a task), answer queries (learn a fact), authority queries (decide with expertise). Everything that fails the 20-Line Test goes in the first bucket.
- Mark the first bucket exit-only. No new investment. Harvest remaining traffic, but budget it as declining by design — Gartner's projection of a 25% structural decline in traditional search volume is the optimistic scenario for pure-utility plays.
- Rebuild around your exclusive inputs. Whatever your organization has that no prompt can synthesize — proprietary data, live systems, lived case work — that's the content engine now.
- Check the moat map quarterly. Canvas's limits shrink every release. What passes the test today by needing an API call may pass into the SERP next year. The 20-Line Test is not a one-time audit; it's a standing early-warning system.
Google just demonstrated, with twenty lines, that it no longer needs to send you the click — only the bill. The sites that survive the action layer will be the ones that made themselves impossible to generate.
Next Blog Post: "Zero-Citation Queries: The Death Map of AI-Era SEO"
Mercury Technology Solutions: Accelerate Digitality.