Guide · Published November 14, 2026 · Reviewed November 14, 2026 · Ingenious AI team · Not legal advice
Colorado ADMT Pre-Use Notice Template: What the Notice Must Say, Line by Line
The short answer: Colorado's SB 26-189 pre-use duty, from January 1, 2027, is a clear and conspicuous notice at the point where the person interacts with the tool, telling them an automated decision-making tool is being used. The notice the kit generates adds the working parts a person needs to act on it: the business name, the tool name, the kind of decision, the correction contact, and the human-review contact. This guide walks the template line by line, using the kit's own document builder as the reference text.
Check your duties and dates — free Build your first notice — free
What is a pre-use notice? It is the notice given before an automated decision-making tool is used on a person — before the screening runs, the score is produced, or the ranking happens. 'Pre-use' is the whole point: the same words delivered with the outcome are a different (and later) document, the adverse-outcome explanation. Colorado's checker result states the duty as a clear and conspicuous notice at the point of interaction; the kit's rules data adds that it names the tool and the kind of decision it informs.
Line one: who is speaking, and which tool
The kit's document builder (app/documents.py) opens the notice with the business and the tool: '[Business] uses [tool] to help make a [decision] decision that may affect you.' Three facts in one sentence, and each is load-bearing. The business name establishes who owes the duty — the deployer, not the vendor. The tool name satisfies the rules data's requirement to name the tool. The decision kind tells the person what is actually at stake for them.
Generic openings fail all three tests at once. 'We use technology to improve our process' names no business duty, no tool, and no decision. If your current draft could be pasted onto any company's website unchanged, it is not yet a Colorado pre-use notice — it is a privacy-policy sentence. The full Colorado duty set around this notice is laid out in Colorado AI Act Automated Decision Notices: What to Send.
The 'what this means' lines: the tool's role, honestly stated
The template's next block states the tool's role: the tool gives information that a person at the business reviews, and its output is one input to the decision; and it states the consequence plainly — the decision can affect the person's access to the opportunity, service, or terms in that process. Both halves matter. The role sentence keeps the notice truthful about how much weight the tool carries; the consequence sentence is what makes the notice worth reading.
Truthfulness here is a factual discipline, not a drafting style. If in your workflow the tool's score is effectively decisive — if reviewers approve its recommendation in nearly every case — then 'one input among many' is the wrong sentence, and the fix is to describe the role as it actually is. The California checker applies the same honesty test from the other direction: token review, where a person rubber-stamps the output, does not make decision-making human. That test is explored in Token Human Review: Why Rubber-Stamping Fails the Test.
Worked example: Front Range Family Dental
Example — fictional business, real duty structure. Front Range Family Dental is a fictional dental group whose hiring platform screens applicants for front-desk roles in Colorado. It adopts the kit's pre-use notice for one hiring round:
- Notices produced: 64 pre-use notices in the round — one per applicant, generated before the screening tool runs on that applicant, each naming the group, the tool, and the hiring decision at stake.
- Contacts carried: every notice prints the office manager as review contact and the HR inbox as correction contact, pulled from the business profile — no notice goes out with a placeholder.
- Corrections triggered: 3 applicants use the correction contact before decisions are made; two files are corrected in the tool's inputs and the corrections are logged against the notices.
- Reviews requested: of 41 applicants not advanced, 5 ask for human review from the contact on their notice; each review is opened from the notice record, not from a fresh search.
- Record kept: 64 notices, 3 corrections, and 5 review requests sit in one chain with the hiring decisions, held for three years.
The rights lines: correction and review, with named contacts
The template then states the person's rights under the state law it names: they can ask what data about them the tool used and ask for corrections, with the correction contact printed on the notice; and if the decision goes against them, they can ask for a plain-language explanation and for review by a person, with the review contact printed. Two rights, two contacts, both on the face of the notice — a right without a contact is a statement, not a path.
In the kit's data model these contacts come from the business profile — a review contact name and email, and a correction contact — so every generated notice carries the current contacts rather than whatever was true when a PDF was last exported. When a contact changes jobs, one profile edit updates the next notice. The correction workflow those contacts feed is detailed in How to Handle ADMT Data Correction Requests.
Placement: clear, conspicuous, at the point of interaction
A perfect text in the wrong place still fails the duty as the checker states it: the notice must be clear and conspicuous, at the point where the person interacts with the tool. Point of interaction means the screening flow, the application step, the assessment start — wherever the person meets the tool. A link in a website footer, a paragraph in a general privacy notice, or a mention in onboarding paperwork the person signs after the tool has run are all placements that ask the person to find the notice after it stopped being useful.
The kit's free flow shows the intended order: the notice is generated for a named person and a named decision before the decision workflow proceeds, and the generation itself is logged. You can test that order yourself in the free notice generator — it produces the same document shape the account side uses, from the same builder. Placement patterns and failure modes get a fuller treatment in Notice Placement: Clear, Conspicuous, at the Point of Interaction.
What the template deliberately leaves out
The generated notice ends with the kit's standard footer: generated with ADMT Notice Kit by Ingenious AI; the kit is a templates and records tool, not legal advice; it records what was said and when; it does not judge whether a decision was fair. That footer is not boilerplate modesty — it marks the boundary of what the document claims. The notice does not promise the person a particular outcome, does not characterise the tool as fair or unbiased, and does not cite duties the state has not imposed.
Equally deliberate: the Colorado notice does not borrow other states' numbers. It does not mention California's opt-out (a California duty with its own exceptions) or Connecticut's 45-day opt-out answer clock. Each state's notice speaks that state's duties. Generate yours against your state in the free duties checker first, so the template you customise is the right state's template.
- Open your current notice draft and check the first sentence names your business, your tool, and the kind of decision
- Rewrite the role sentence until it describes the tool's actual weight in your decision — not the weight you wish it had
- Put a named correction contact and a named review contact on the face of the notice, from one maintained profile
- Move the notice to the point of interaction: before the tool runs on the person, in the flow they are actually in
- Generate a test notice in the free generator and read it as an applicant would — then log every real notice you send
Sources, as carried in the kit's own data: the duty summaries in app/rules.py and the checker data in app/checker_data.py, verified against primary or authoritative sources on October 28, 2026. Run the free duties checker for your jurisdiction, decision type, and the tool's role, and compare the kit's plans for the record, builder, and review queue. More from Ingenious AI is on the hub.
Related guides
- Colorado AI Act Automated Decision Notices: What to Send
- Token Human Review: Why Rubber-Stamping the Tool's Output Fails Every State's Test
- How to Handle ADMT Data Correction Requests: From Intake to Re-Run Decision
- Notice Placement: Clear, Conspicuous, and at the Point of Interaction
See all ADMT guides for the full set.
Frequently asked questions
What must a Colorado pre-use notice say?
Per the kit's data: a clear and conspicuous notice at the point of interaction that an automated decision-making tool is being used. The kit's template adds the business name, the tool name, the kind of decision, the correction contact, and the human-review contact.
When must the Colorado pre-use notice be given?
Before the tool is used on the person — at the point where they interact with it. A notice delivered with the outcome is not a pre-use notice.
Does the notice have to name the tool?
Yes — the kit's Colorado rules data states the notice tells the person what kind of decision the tool informs and the tool name.
What contacts belong on the notice?
A correction contact (for asking what data was used and correcting it) and a human-review contact (for requesting explanation and review after an adverse outcome). The kit pulls both from the business profile.
Is the Colorado pre-use notice the same as California's?
Similar shape, different duties. California's pre-use notice (section 7220) describes the use and names the opt-out and access rights; Colorado's centres on the point-of-interaction notice, correction, and the 30-day explanation. Each state's template speaks its own duties.