Free plan

← All guides

data_use_terms Explained: Retention, Resale, and Training

Published November 14, 2026 · Reviewed November 2026 · By the Ingenious AI team

The short answer

data_use_terms is the optional per-rule field in an agent-policy.json file where the owner states, in plain language, what an agent may do with the site's data after it fetches it: retain it, resell it, train models on it — or none of those. It is free text, not a code, because data use is too varied for three tidy values. The store sample's version is the model: "Product catalog data may be used for price comparison and shopping assistance only. No resale of catalog data. No model training on product images."

Generate your policy — free · Validate an existing one

What are data-use terms?

They are the after-life clause of a rule. The action lists govern the visit — what the agent may do while it is on your site. Data-use terms govern everything after the visit: the copies, caches, indexes, and training sets your content might land in. A rule can permit read generously and still refuse the after-life — read to answer a question now, but do not keep the catalogue, do not resell it, do not train on it. null, the field's other state, states no terms: the owner has expressed no preference, which agents should read as silence, not permission for anything in particular.

The three subjects worth addressing

Nearly every useful data-use sentence covers some of three subjects. Retention: how long fetched data may be kept — "retain for the duration of the user's task", or "cache prices for up to 24 hours", both of which suit price data that rots. Resale and redistribution: whether your data may be passed on, repackaged, or sold — the clause catalogue owners care about most, since a permitted read at 60 requests a minute plus silent resale is a competitor's feed wearing your permission. Training: whether content may be used to train models — the clause publishers reach for first, often as a flat "no model training on article text or product images". Write the subjects that apply to your rule, in sentences, and skip the rest; a short true clause beats a long borrowed one.

Per rule, because data differs by agent

The field sits inside each rule, so the same site can offer different after-lives to different agents. The store sample does: its shopping rule limits use to price comparison and shopping assistance, while its reading rule requires that quotes link back to the source. A publisher might let a named research agent retain article metadata indefinitely for an index, while the catch-all gets "no retention beyond the session". This granularity is the field's reason to exist — and its discipline. Copy-pasting one boilerplate sentence into every rule wastes it; the sentence should change exactly where the bargain changes, alongside the commercial terms it complements.

Writing sentences a machine era can use

The readers of this field are agent operators and, increasingly, the agents' own planners summarising terms for their users. Write for both: short declarative sentences, named prohibitions, no legal fog. "No resale" survives summarisation; "data shall not be utilised for purposes extraneous to the contemplated engagement" does not. The honest-scope caveat applies with full force here — the spec calls the whole policy a unilateral declaration, not legal advice, and data-use terms are its least mechanical part. They state your position in the standard place, in the standard company of the verdicts and limits around them. That stating, done plainly and kept current, is what the field is for. Pair it with the right actions: refusing training means little if crawl-full-site is permitted without comment, a pairing worked through in the publishers' guide.

There is a temptation, drafting this field, to borrow the longest legal sentence available and paste it into every rule as armour. Resist it on operational grounds alone: an agent's planner summarising your terms for its operator will compress whatever you write, and compression of legal fog produces confident nonsense — a summary that grants what you refused, delivered in your name. The plain sentences this guide recommends are not a stylistic preference; they are the only form in which your intent survives the channel the field actually travels through. Where a clause genuinely needs legal precision, put the plain version in the rule and the precise version in the document the rule's readers can be pointed to — precision by reference, clarity in place.

Worked example: a recipe publisher writes three different after-lives

Fictional example. Saffron & Smoke, a fictional recipe publisher with 2,300 recipes, writes three distinct data-use sentences across its rules. Shopping agents — permitted to read and shop for ingredient lists — get: "Ingredient and price data may be used for shopping assistance only. No retention beyond 7 days." A named meal-planning agent gets a wider grant: "Recipes may be retained and indexed for meal planning. Servings and ingredients only — no republication of method text." The catch-all gets the short refusal: "No retention, resale, or model training." When a fourth agent's operator wrote in asking to license the full corpus, the conversation started from the published sentences — each side could point at the exact clause it wanted to change.

Writing data-use terms in five steps

  1. Name the after-life. Decide what may happen to your data after the visit, per rule.
  2. Cover retention. Say how long fetched data may be kept, or that it may not be kept.
  3. Cover resale. State plainly whether data may be resold or redistributed.
  4. Cover training. Say whether content may be used to train models.
  5. Keep it per-rule. Let the sentence change wherever the bargain changes — boilerplate everywhere is information nowhere.

Keep reading

Agent Identifiers Explained: the vendor/agent Format · The generic/* Rule: Your Policy's Catch-All, Explained · Agent Access Policy Examples · Spec samples · Conformance · Pricing · All guides

Frequently asked questions

Is data_use_terms required?

No. It is optional on every rule, and null states that no terms are expressed. Optional is not the same as unimportant: for publishers and catalogue owners it is often the clause they care about most, because it governs what happens to their content after the visit ends.

Can data-use terms contradict the action lists?

They govern different things — actions govern the visit, data-use terms govern afterwards — but a pairing can still be incoherent: permitting crawl-full-site while writing 'no retention' asks an agent to copy everything and keep nothing. The validator cannot catch that; it is a drafting judgement, and reviewers should read the two together.

Should data-use terms be legal language?

Plain language serves better. The field's readers include software and the operators who configure it. Short declarative sentences — what may be used for what, what may not happen at all — survive being parsed, summarised, and quoted back to you. The spec states the policy is not legal advice in any case.

What does the blog sample put in its data-use terms?

Attribution. The blog that is open with attribution welcomes agents that attribute and link back, and refuses bulk scrapers that copy full articles. Its terms show the field's range: a condition about credit, not price — data-use terms carry whatever non-price conditions matter to the owner.

Can I update data-use terms without changing anything else?

Yes. Edit the sentence in the rule, update last_updated, and re-validate. The field's independence is practical: data bargains evolve faster than access verdicts, and the format lets you revise the after-life clause without reopening who may visit.

Generate a policy — free · Validate a policy · What an agent access policy is

← Back