The free AI tier has a price: Google's own pricing page lists 'content used to improve our products' as a Free-plan feature

# Your free AI tier has a price, and it is printed on the pricing page

Open Google's Gemini API pricing page and look at the Free plan's feature list. It reads like any other feature list. Limited access to certain models. Free input and output tokens. Google AI Studio access. Then, with the same checkmark in front of it as all the others:

> Content used to improve our products

That is not a line from a privacy policy. It is a selling point, sitting in the column you choose *because* it costs nothing, formatted exactly like the things you are being offered rather than the thing you are paying with.

One column to the right, the Paid plan carries the same row inverted — *Content not used to improve our products* — and further down, the per-model table drops the sales voice entirely and just says it:

For Gemini 3.8 Flash, the free tier's input and output are both listed as "Free of charge." The paid column for that same model is $0.75 per million input tokens and $3.75 per million output, through 31 December 2026; both double on 1 January 2027. For scale, Anthropic lists Claude Sonnet 5 at $2 in and $10 out per million, and OpenAI lists GPT-5 at $1.25 and $10.

So the row that answers No is not a premium tier or an enterprise contract. On a side project that never clears a million tokens, it is small change. The reason most people are on the other side of that row is not that they weighed it and chose. It is that nobody reads a feature list looking for the feature that is about them.

Short-form is currently full of advice to get your AI for free — free API directories, free model tiers, permanent free access. That advice is not wrong. Free tiers are real, they are genuinely useful, and for throwaway prompts they are the correct choice.

But "free" and "unpriced" are different words, and only one of them is true here. The trade is disclosed. It is disclosed in the least alarming place imaginable, which is why almost nobody has read it.

Worth adding: one repo currently circulating as a directory of free APIs advertises "134 free APIs from over 40 providers." Its own README says 453 models from 31 providers, and its own code examples say 128. When a source cannot agree with itself about its headline number, that is worth knowing before you paste a key from it.

Nothing dramatic, and nothing that requires switching providers.

Use the free key for prompts you would not mind reading aloud. Scratch tests, throwaway questions, anything synthetic. This is what a free tier is genuinely good at and there is no reason to stop.

Use a paid key for anything that is yours. Your own source, a client's data, an unreleased idea, anything under an NDA. The switch is an environment variable and the bill is cents.

In practice that is three lines, and the middle one is the whole change:

pip install google-genai                    # the official SDK; it is also what the checker looks for
export GEMINI_API_KEY="<your paid key>"     # this is the switch — the column that answers No
python3 tier-check.py .                     # what this project is actually wired to

It will not undo what you have already sent. It changes what happens to the next thing.

tier-check.py is stdlib-only Python that walks a project, finds which providers it actually talks to — env var names, SDK imports, API hosts — and prints each one's published free-tier posture with the URL it came from. It strips every value to the right of an = before matching, so it can tell you GOOGLE_API_KEY is referenced without ever touching what it contains. It sends nothing anywhere and needs no key of its own.

It cannot see which tier your key is on. Only your provider console knows that, and the kit says so on its own first page. Two providers' rows are marked NOT ESTABLISHED rather than guessed, because their rate-limit pages render client-side and could not be read from a primary source on the day this was written. A row that says "go and read it" beats a confident number that is wrong.

All figures above were read directly from the vendors' own pages on 7 September 2026. Vendors change them. The links are there for that reason.