Free AI tiers removed the cost of writing. They did not remove the cost of being wrong at scale. If a model can draft ten posts an hour for nothing, the only thing standing between you and a site full of pages Google refuses to index is a rule that tells you when to stop.
This is the rule we run: publishing pauses when the indexed share of mature posts drops below 60%. Below is how the gate works, what our own Search Console snapshot looked like at 37 live posts, and how to build the same check without writing code.
How do you use AI for free without publishing junk?
Use free AI tiers for drafting, but gate publishing on indexation instead of output. Track the share of your mature posts that Google actually indexes, and pause new posts when that share falls below 60%. Free generation removes the cost of writing; it never removes the cost of a diluted site.
The mistake is treating “free” as a reason to publish more. Free tiers change your marginal cost, not Google’s willingness to index you. Google states plainly in its Search Essentials that indexing is not guaranteed for any page, and its spam policies added “scaled content abuse” in March 2024 — targeting mass-produced pages made primarily to manipulate rankings, regardless of whether a human or a model wrote them.
Ahrefs’ study of roughly one billion pages found that 90.63% of pages get no organic search traffic from Google at all (Ahrefs, 2020). That is the baseline you are publishing into. Volume alone does not move you out of it.
What is the 60% index-rate gate?
It is a throttle: before each scheduled publish, the engine calculates what percentage of already-published, mature posts Google has indexed. If that number is at or above 60%, publishing proceeds. If it is below 60%, publishing stops until the backlog clears. The gate measures absorption, not production.
The formula is deliberately simple:
- Mature posts = posts old enough that Google has had a fair chance to crawl them. New posts are excluded, or a fresh publishing run would drag its own score down.
- Indexed = counted from Search Console’s Page Indexing report, not from a
site:search, which is unreliable for coverage checks. - Gate = indexed ÷ mature. Below 0.60, the scheduler skips its run and logs why.
The maturity window is a choice, not a constant. Ahrefs’ study of newly published pages reported that only 5.7% of them reached Google’s top 10 within a year — indexing is faster than ranking, but neither is instant. Pick a window long enough that a normal crawl would have happened, and keep it fixed so the number stays comparable week to week.
What did our Search Console snapshot look like at 37 posts?
At 37 live posts, our Search Console snapshot showed roughly 79% indexed — comfortably above the gate. This is a self-reported reading from one site on one day, not a benchmark or a controlled test: treat it as an illustration of how to read the report, not as a number to match. Two categories accounted for much of the non-indexed remainder — 5 URLs sitting in “Discovered – currently not indexed,” and 2 URLs still reported as submitted long after we had deleted them from the site.
Those two stale URLs matter more than they look. They mean the reported denominator and the real denominator disagree. If you calculate a gate percentage without reconciling deleted pages, you are grading yourself against a site that no longer exists.
What do those Search Console statuses actually mean?
“Discovered – currently not indexed” means Google knows the URL exists but has not crawled it, often because crawling it would overload the site or because Google has not judged it worth the request. “Crawled – currently not indexed” means Google fetched the page and declined to index it — a quality signal, not a capacity one.
| Status | What Google is telling you | What the engine should do |
|---|---|---|
| Indexed | Page is in the index and eligible to rank | Counts toward the gate numerator |
| Discovered – currently not indexed | URL known, not yet crawled (5 of ours sat here) | Wait; check internal links and sitemap freshness before adding volume |
| Crawled – currently not indexed | Fetched and rejected — usually thin or duplicative | Treat as a content-quality signal; stop generating that cluster |
| Duplicate, Google chose different canonical | Another page already covers it | Merge or delete; this is cannibalization made visible |
| Submitted URL not found (404) | Sitemap still advertises a deleted URL (2 of ours) | Regenerate the sitemap; exclude from the denominator |
Definitions here follow Google’s Page Indexing report documentation. Read the statuses as instructions, not as errors to dismiss.
Volume-first or index-gated: what actually changes?
Volume-first publishing optimizes for posts shipped. Index-gated publishing optimizes for posts Google accepts. The output looks identical on day one and diverges sharply by month three, because a volume-first site keeps adding to a pile Google has already declined to crawl, while a gated site pauses and fixes the cause.
| Dimension | Volume-first (free-tier default) | Index-gated |
|---|---|---|
| Success metric | Posts published per week | Indexed share of mature posts |
| Response to a bad week | Publish more | Publish nothing until the ratio recovers |
| Failure mode | Silent dilution; nobody notices until traffic never arrives | Visible stall; the log states why it paused |
| Cost of AI drafting | Near zero, so nothing restrains it | Near zero, but the gate restrains it |
| What it protects | Cadence | Crawl budget and topical clarity |
How do you build the gate step by step?
You do not need to code it. You need a repeatable weekly calculation, a written threshold you agreed to before you saw the number, and a scheduler that will actually refuse to run. The hard part is not the arithmetic — it is committing to a rule that will sometimes tell you to publish nothing.
- Export the Page Indexing report and count indexed versus non-indexed for posts older than your maturity window.
- Remove deleted URLs from both sides of the ratio, then regenerate the sitemap so the stale entries stop reappearing.
- Compute the ratio and write it down with the date. One number, one line, every week.
- Set the gate at 60% and make the scheduler read it before publishing.
- When it trips, fix causes — merge duplicates, add internal links to orphaned posts — rather than lowering the threshold.
What else has to be blocked before generation?
Duplicate topic clusters, checked before a draft exists. Our engine rejects candidate topics that cannibalize posts already published, and on one run it rejected 7 of 7 candidates. That was not a bug — it was the practical signal that the seed keyword list for that niche had been exhausted and needed replacing, not re-running.
This matters because free AI will happily generate the eighth post about the same thing. The block belongs upstream of generation, where rejecting a topic costs nothing, rather than downstream in Search Console, where you learn about it as “Duplicate, Google chose different canonical.”
What can go wrong with the gate itself?
The gate can fail while reporting success. Our verification step died twice — once on authentication, once on a usage quota — and publishing dropped to zero for 24 days, then 3 days, while every scheduled task still reported success. A gate that cannot fail loudly is not a gate; it is a decoration.
Three fixes, all learned the hard way by someone who cannot read code and diagnosed each failure by pasting the error back into the AI:
- Split the check across two vendors. A single verification provider is a single point of silent failure.
- Alert even when the fallback succeeds. Silence must mean “working,” never “the alarm also broke.”
- Keep the watchdog outside the system it watches. A monitor that dies with the process it monitors reports perfect health forever.
What this approach does not tell you
Index rate is a health check, not a traffic forecast. A post can be indexed and still earn nothing, and 60% is our threshold on our site — not a published Google standard. Treat the number as a stop signal for volume, and judge value separately by clicks and impressions.
The honest summary of how to use AI for free: let the model make drafting cheap, then spend the savings on a rule that limits how much of it reaches the web. Ours pauses at 60%. Yours should exist, be written down before the number is inconvenient, and be loud enough to hear when it breaks.