Initial CTA

Exemplar page — first pass, entirely proposal. No Invoca product ships this pattern today. Nothing below is code fact or measured behavior — it is a proposal offered for review. See Coverage, stated honestly.

What it is

A call center manager opens a call record that has never been summarized. The transcript is long; nothing on the page says a summary is available until an Initial CTA does — a labeled invitation, distinct from the surrounding content, offering to run Signal AI against this specific transcript. The manager did not go looking for an AI feature. The CTA is what tells them one exists here, right now, for this record. The same job recurs anywhere an agentic feature sits next to work a user already knows how to do manually: a marketer staring at a filter panel who could instead type a question in plain language; a report view that could be drafted rather than assembled by hand. In every case the user’s default path does not require the AI feature — the Initial CTA is the entire reason they find out it is there. An Initial CTA is not a standing menu item and not a suggestion attached to existing AI output. It fires once, for a user who has not started, at the specific place in the product where starting would help. Once that user has engaged — has an output on screen to react to — the next nudge is a Suggestion, not another Initial CTA.

Choose this when / choose something else when

Situation	Do this instead	Why
The user already has AI output on screen and needs a nudge toward what’s next	Suggestions	Initial CTA is for someone who hasn’t started; Suggestions assumes an existing context to build on
The invite is really about changing scope, tone, or depth before a run, not whether to start one	Controls	Tuning behavior is a different job than the first invitation to use the feature at all
The control’s job is to navigate to a separate page or URL rather than trigger the feature in place	Link	Per TITAN-BTN-08, a control that navigates to a URL is a Link, not a Button, whether or not AI is involved
The action being invited is not AI-backed at all	Button	This page only governs the entry point to a model-generated result; a deterministic action is an ordinary Button with no disclosure obligation

Agency tier

Suggests. Pressing an Initial CTA proposes running the feature — it does not commit the user to anything and produces output the user reviews, the same as any other Suggests-tier surface. The CTA itself performs no action beyond starting the request. This would have to change if the CTA were wired to run and immediately apply a result without a review step — at that point the tier is no longer the CTA’s to state; it inherits whichever tier the triggered Action carries, and that escalation needs to be recorded, not discovered.

Anatomy

┌─────────────────────────────────────────────┐
│  ✨ Summarize this call                      │
│  Signal AI can generate a summary from       │
│  this transcript in about 10 seconds.        │
│                                               │
│                    [ Not now ]  [ Summarize ]│
└─────────────────────────────────────────────┘

#	Part	Component	Required
1	Container	Card	Yes
2	AI identifier mark (icon and/or name)	See Identifiers	Yes — see Disclosure & recourse
3	Invitation copy — what will happen and roughly how long	Text content, no component	Yes
4	Trigger control	Button, `primary` prominence	Yes
5	Dismiss control	Button, `tertiary` prominence	No

At most one Initial CTA is primary in a given region, per TITAN-BTN-01 — the invitation is one specific offer, not a menu of equally weighted options.

Outcome states

State	Treatment
Working	Trigger button enters `loading`, preserving width, once pressed
Streaming / partial	Not applicable to this page. The CTA has no output of its own; once triggered, partial output belongs to the invoked Action’s own page.
Confident and right	Not applicable. The CTA carries no result to be right about.
Confident and wrong	Not applicable — by design, not by omission. The CTA never renders a generated answer; it only offers to start. The “confident and wrong” case is fully the invoked Action’s to design, not this page’s.
Uncertain	Not applicable, for the same reason as above.
Refused	The feature declines to run — e.g., the transcript is too short to summarize usefully. State why inline, next to the CTA, not as a separate error.
Empty	Nothing to run against — e.g., no transcript exists yet. The CTA does not render at all rather than rendering disabled with no explanation.
Interrupted	User navigates away after pressing but before a result returns. No output exists to preserve; next visit shows the CTA again, not a stuck loading state.
Degraded	The feature is available but running on a fallback (a cached or rule-based result instead of a live model call). Disclosed in the invitation copy itself, not discovered only after pressing.
Rate-limited / over budget	The CTA renders disabled with the reason and, where known, when it will be available again — never silently missing.
Stale	Not applicable. Nothing has been generated yet for staleness to apply to.

Disclosure & recourse

Does the user know this is AI, at the moment it matters? Yes — the identifier mark and name are part of the CTA itself, not disclosed only after pressing.
What did it use? State it in the invitation copy: this call’s recording and transcript, nothing else, unless something else is genuinely used.
How sure is it, and does that change behavior? Not applicable at this stage. No output exists yet to be sure or unsure about; confidence is the invoked Action’s concern.
How does the user check it? Not applicable yet, for the same reason.
How does the user correct it? Not applicable yet.
How does the user get out? Dismissing the CTA starts nothing and costs nothing — no run occurs, no record of the dismissal changes what the user sees next time unless the product deliberately reduces frequency. The CTA reappearing every visit is preferable to a silent permanent opt-out.

Reference

No model, prompt, tool schema, latency budget, or cost has been defined for this pattern. Where the CTA states “about 10 seconds,” that number is illustrative copy for this page, not a measured latency.

Evaluation

Not evaluated. No eval set exists for this pattern or for any pattern it invokes, except Signal AI’s shipped transcript summary, whose evaluation status is recorded on its own page.

Content

Element	✅	❌
Invitation title	Summarize this call	Try AI!
Invitation body	Signal AI can generate a summary from this transcript in about 10 seconds.	Let our advanced AI help you save time!
Trigger button	Summarize	Go
Dismiss control	Not now	No thanks

Follows Voice and tone: plain language, no exclamation, no adjectives doing the button’s job. State what will happen and how long it takes; do not sell it.

Accessibility

The trigger button’s loading state is announced via aria-live="polite" on a region containing the button, not by the button’s visual state alone — a screen reader user needs to know a request is in flight without waiting on the eventual result to discover it retroactively.
An indeterminate wait (before a duration estimate is known) needs a non-visual equivalent to a spinner — an accessible status text (“Working…”), not only a visual animation.
If the CTA is disabled for a rate limit or an unmet precondition, the reason is in the accessible description (aria-describedby), not conveyed by disabled styling alone — see TITAN-COLOR-03.
Focus stays on the trigger button through the loading state and moves to the result (or to an inline refusal message) only once one exists — never left stranded on a button that has already been pressed and dismissed from view.

Constraints

ID	Constraint	Rationale
TITAN-CTA-01	An Initial CTA states, in its own copy, that the result is AI-generated. Disclosure does not wait for the result to appear.	TITAN-AI-04: the user learns this is AI at the point of the output, not only afterward — the CTA is the earliest point available.
TITAN-CTA-02	An Initial CTA states what will happen and, where known, roughly how long it takes, before the user presses it.	An invitation with no stated outcome asks for trust with nothing to evaluate it against.
TITAN-CTA-03	At most one Initial CTA renders as `primary` prominence within a given region.	TITAN-BTN-01 — an invitation competing with another equally weighted control is not an invitation, it’s a choice the interface refused to make.
TITAN-CTA-04	Dismissing an Initial CTA never permanently removes it without the user taking a separate, explicit action to turn the feature off.	A one-time dismissal that silently becomes a permanent opt-out removes recourse the user did not know they were giving up — see TITAN-AI overview, question 6.
TITAN-CTA-05	A degraded fallback (cached or rule-based result instead of a live model call) is disclosed in the invitation copy before the user presses, not only in the result afterward.	A result that looks the same regardless of how it was produced hides a real difference in what the user should trust.
TITAN-CTA-06	An Initial CTA that cannot run right now (rate-limited, precondition unmet) renders disabled with the reason stated, never simply absent with no explanation.	An absent control teaches the user nothing; a disabled one with a reason teaches them when to come back.
TITAN-CTA-07	The CTA never uses a Button for what is actually navigation to a separate page.	TITAN-BTN-08 applies regardless of whether the destination is AI-backed.

Divergences

Not applicable — nothing is shipped yet to diverge from.

Gaps

Whether an Initial CTA should ever re-appear after a user has already used the feature several times, or whether it should permanently give way to ambient access (a persistent icon, a menu entry) is undecided.
How “roughly how long it takes” is computed, and whether it is shown at all when latency is unpredictable, is undecided.
Whether dismissal frequency should decay (show every time, then less often) is undecided — see TITAN-CTA-04, which sets the floor (never a silent permanent opt-out) but not the specific curve.
Whether an Initial CTA can appear more than once per page for different, unrelated AI features, and how those are visually distinguished from each other, is undecided.

Volatility

This page’s guidance depends on no model capability directly — it governs the invitation, not the output — but it assumes an agentic feature can state, before running, roughly what it will do and how long it will take. If a future feature cannot estimate its own latency or cannot describe its output in advance, TITAN-CTA-02 needs reverification against that case. Dated 2026-09-02; revisit on the first real product implementation of any Wayfinder.

AI Experience overview — vocabulary, agency tiers, and the six disclosure questions this page answers
Suggestions — the next nudge, for a user who has already started
Open input — often what an Initial CTA opens into, for features that need the user to say what they want
Controls — for adjusting behavior rather than inviting a first run
Actions: Summarize — Signal AI’s shipped job, the concrete case this page’s examples are drawn from
Button, Card, Link — the components this pattern composes

Why it works this way

The CTA’s entire job is to make the default path optional, not to replace it. A user who ignores it can still do the work manually — read the transcript, build the filter by hand — so the CTA’s only leverage is stating clearly enough what it offers that trying it costs less than it might save. That is why disclosure and a stated outcome sit before the button rather than after: an invitation asking for trust with no way to evaluate it is a worse deal than the manual path it is trying to replace.

Overview

Wayfinders

Inputs

Tuners

Governors

Trust builders

Identifiers

Actions

What it is

Choose this when / choose something else when

Agency tier

Anatomy

Outcome states

Disclosure & recourse

Reference

Evaluation

Content

Accessibility

Constraints

Divergences

Gaps

Volatility

Why it works this way

​What it is

​Choose this when / choose something else when

​Agency tier

​Anatomy

​Outcome states

​Disclosure & recourse

​Reference

​Evaluation

​Content

​Accessibility

​Constraints

​Divergences

​Gaps

​Volatility

​Related

​Why it works this way

What it is

Choose this when / choose something else when

Agency tier

Anatomy

Outcome states

Disclosure & recourse

Reference

Evaluation

Content

Accessibility

Constraints

Divergences

Gaps

Volatility

Related

Why it works this way