Welcome / empty state

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

This page covers two moments, both specific to an agentic feature and both distinct from a plain empty state: Welcome is what a marketer sees the first time they open an AI panel that hasn’t been asked anything yet — a natural-language search over call data, or a summarization panel on a call that hasn’t been analyzed. There is no prior run to show, so the panel’s only job is to set expectations: what this feature can do, what it can’t, and how to start. A user who doesn’t yet know that Signal AI summarizes transcripts but can’t change routing or settings needs to learn that here, before they ask for the wrong thing and get refused. No-results is what the same marketer sees after they’ve asked something — “show me calls about a refund dispute from last week” — and the search, summarize, or recommend run completed correctly and found genuinely nothing. The run worked. There’s simply nothing to show. That distinction — worked-but-empty versus broken, loading, or refused — is the entire content of this half of the page, and it’s the one a plain empty state doesn’t have to carry, because a deterministic region either has data or it doesn’t, with no “the search itself might have been wrong” question attached. Both moments are visually related — both replace the same panel’s content with a message plus optional next step — and both are covered on this page because both carry the same AI-specific obligation a static empty region never has: telling the user something about the feature’s scope, not just about the current state of the data.

Choose this when / choose something else when

Situation	Do this instead	Why
The region has nothing to show and no AI feature is involved	Empty and zero states and EmptyState	Those pages cover the deterministic case: a table with no rows, a list with nothing created yet. Neither needs to set capability expectations, and neither risks being confused with “still thinking” — a deterministic empty region was never going to produce a result on its own. This page exists because an AI panel’s emptiness carries an extra question a plain one doesn’t: did the run actually finish, and does the user understand what it can do at all?
A run is still in progress	The Working treatment — a Skeleton or Spinner	This pattern renders only once a run has definitively finished with nothing, or before one has started. Showing it while a run is still going is the single most damaging mistake this page exists to prevent — see TITAN-AIWELCOME-01.
The request had more than one plausible reading, not zero results	Disambiguate	Ambiguity and emptiness are different problems — one has too many candidate answers, the other has none.
There’s nothing notable to recommend this period	Recommend’s own Empty outcome state	Recommend’s empty case composes this pattern’s no-results treatment; document the copy specifics there, not by duplicating this page.

Agency tier

Not applicable — this isn’t an acting surface at all. Welcome and no-results are the moment before a run starts or the moment after one finishes with nothing; nothing is proposed, drafted, or acted on in either state, so there’s no authority to name. The agency tier belongs to whatever the user does next — the Search, Summarize, or Recommend run this panel leads into.

Anatomy

Welcome:

┌─────────────────────────────────────────────┐
│                    ◆                          │
│           Ask about your calls                │
│   Signal AI can summarize transcripts and      │
│   surface Signals. It can't change routing     │
│   or settings.                                 │
│                                                 │
│   [ "Show me calls about billing" ]            │
│   [ "Summarize today's calls" ]                │
│                                                 │
│   [______ Ask a question ______________] [→]  │
└─────────────────────────────────────────────┘

No-results:

┌─────────────────────────────────────────────┐
│              (illustration)                    │
│   No calls matched "refund dispute"            │
│   Try a broader date range, or rephrase        │
│   your question.                               │
│                                                 │
│      [ Rephrase ]        [ Clear filters ]     │
└─────────────────────────────────────────────┘

#	Part	Component	Required
1	AI identifier mark (welcome only)	See Identifiers	Yes, for welcome
2	Illustration (no-results only)	EmptyState’s `search` illustration	Yes, for no-results — never the loading or error illustration
3	Heading	Text content, no component	Yes
4	Scope-setting line — what it can and can’t do (welcome only)	Text content, no component	Yes — see TITAN-AIWELCOME-02
5	Example prompts (welcome only)	Tag or Button `outlined`, clickable	Recommended
6	Input	Input	Yes, for welcome
7	Next-step actions (no-results only)	Button	Recommended — see TITAN-EMP-05

The no-results variant composes EmptyState directly — same illustration-plus-title-plus-supporting-line shape that component already defines. What’s added is the AI-specific obligation: the heading names what was searched for, in the user’s own words, not a generic “no results.”

Outcome states

State	Treatment
Working	Not applicable to this pattern — by construction. This pattern renders only before a run starts or once one has definitively finished. A run in progress is a Skeleton or Spinner, never this pattern’s copy or illustration.
Streaming / partial	Not applicable, for the same reason — nothing here is a partial result.
Confident and right	Not applicable. No output was produced in either the welcome or no-results case; there’s nothing to be right about.
Confident and wrong	The no-results claim is itself a claim, and it can be wrong. A search that says “no calls matched” when a manual filter over the same data turns up results is a confidently wrong empty state — the failure looks identical to a correctly empty one. The user notices only by trying the equivalent manual path (a plain filter, a direct table view) and finding what the AI search missed, or by already knowing the data should be there. Nothing about the no-results screen itself signals that the search might have missed something — see TITAN-AIWELCOME-04.
Uncertain	If the search was cut short (timeout, partial index) rather than exhaustive, the copy says “the search may be incomplete” rather than a flat “no calls matched” — see TITAN-AIWELCOME-05.
Refused	Not applicable to this page. A declined request (disallowed query, out-of-scope ask) is the invoking action’s own Refused state, documented there — this pattern only covers a request that ran and legitimately found nothing.
Empty	This is the no-results half of this pattern — a run that completed and genuinely found nothing to show.
Interrupted	A cancelled or dropped run must not render as no-results. Show that the run was stopped, not that it finished empty — conflating the two tells the user a search happened when it didn’t.
Degraded	If the empty result came from a fallback path (keyword-only matching instead of full semantic search), the no-results copy discloses that the full search wasn’t run, since the full version might have found something.
Rate-limited / over budget	If the search stopped early because it hit a budget, the copy says “search stopped early” — never “no calls matched,” which claims a completeness the run didn’t have.
Stale	Minor relevance. A cached no-results answer can be stale if new calls have since arrived; where results are cached, show when the search last ran.

Disclosure & recourse

Does the user know this is AI, at the moment it matters? Yes for welcome — the scope-setting copy names the feature and is the disclosure itself. For no-results, the framing (“your question didn’t match anything”) makes clear a query was interpreted, not just filtered.
What did it use? For no-results, state what was searched — which calls, which date range — so “nothing matched” is checkable against a concrete scope, not a vague “your data.”
How sure is it, and does that change behavior? For no-results, an exhaustive search says so plainly (“no calls matched”); an incomplete one says so differently (see Uncertain) and that difference changes what recourse is offered.
How does the user check it? The user can always fall back to a manual filter or the underlying Table view to verify a no-results claim themselves — see Filtering and search.
How does the user correct it? Rephrasing or broadening (no-results) and trying an example prompt (welcome) are the two correction paths. Neither persists as a standing preference.
How does the user get out? The manual, non-AI path — the plain table or filter view — is always available and unaffected by anything this panel did or didn’t find.

Reference

No model, prompt, tool schema, latency budget, or cost has been defined for this pattern.

Evaluation

Not evaluated. No eval set exists for whether “no results” is ever produced incorrectly (a false empty) versus correctly.

Content

Element	✅	❌
Welcome heading	Ask about your calls	AI Assistant
Welcome scope line	Signal AI can summarize transcripts and surface Signals. It can’t change routing or settings.	I can help with almost anything!
Example prompt	”Summarize today’s calls”	Try me!
No-results heading	No calls matched “refund dispute”	No results
No-results supporting line	Try a broader date range, or rephrase your question.	Something went wrong.

Never reuse EmptyState’s generic fallback copy (“No Records found!”) for a no-results moment — it doesn’t name what was searched for and reads as broken rather than as a completed, empty search. Per TITAN-EMPTYCOPY-03, the no-results wording names the query; per this page’s own TITAN-AIWELCOME-02, the welcome wording names what the feature can’t do, not only what it can.

Accessibility

The transition from a loading region to the no-results message is announced via a polite live region on completion — not token-by-token, and not silently, so a screen reader user can tell the difference between “still running” and “finished, found nothing.”
An indeterminate wait before either state renders needs a non-visual equivalent to a spinner — accessible status text, not a purely visual animation.
The scope-setting line and the no-results heading are both real text, never baked into an illustration — matches TITAN-EMP-07.
When a no-results screen replaces a loading region, focus moves to the heading so a screen reader user lands on the result, not left on a control that no longer has anything to act on.
Example prompts (welcome) and next-step actions (no-results) are both reachable by keyboard in visual order.

Constraints

ID	Constraint	Rationale
TITAN-AIWELCOME-01	This pattern never renders while a run is in progress.	Showing “no results” or a welcome message over a search that’s still running tells the user something false about whether their request was heard.
TITAN-AIWELCOME-02	The welcome moment states what the feature can’t do, not only what it can.	A user who only learns the feature’s strengths will ask for something out of scope and get an unexplained refusal instead of a set expectation.
TITAN-AIWELCOME-03	The no-results heading names the specific request that returned nothing, in the user’s words.	”No calls matched ‘refund dispute’” is checkable; “No results” gives the user nothing to verify a mis-scoped search against.
TITAN-AIWELCOME-04	A no-results screen and a run cancelled before completion use visibly and verbally different treatments.	Conflating “found nothing” with “never finished” hides a run that may need to be retried behind a claim that it was exhaustive.
TITAN-AIWELCOME-05	A no-results claim states whether the underlying search was exhaustive or cut short (timeout, fallback, budget).	”No calls matched” implies completeness the run may not have had; overstating it costs the user a result that existed but wasn’t found.
TITAN-AIWELCOME-06	Example prompts on the welcome screen reflect jobs the feature actually does today, never an aspirational capability.	An example prompt the feature can’t yet fulfill teaches the wrong lesson at the first interaction.
TITAN-AIWELCOME-07	The no-results state never reuses a generic fallback empty-state message with no reference to what was searched.	See Content — a message that doesn’t name the query reads as a broken load, not a completed search.

Divergences

Not applicable — nothing is shipped yet to diverge from.

Gaps

Whether the welcome screen’s example prompts should be static or drawn dynamically from what the specific account or user has actually asked successfully before is undecided.
Whether a no-results moment should offer to fall back to a broader, AI-assisted retry (e.g., automatically widening the date range) or only manual recourse is undecided.
Whether “the search may be incomplete” (see Uncertain) needs its own distinct visual treatment from a fully exhaustive no-results screen, or whether the copy difference alone is sufficient, is undecided.
How this pattern’s welcome moment differs, if at all, from an Initial CTA for the same feature is not fully resolved — the working distinction on this page is that Initial CTA is a one-time invitation attached to a specific piece of content, while welcome is the standing state of a panel with nothing asked yet, but the boundary hasn’t been tested against a real product surface.

Volatility

This page assumes an agentic feature can distinguish, internally, between “ran and found nothing” and “didn’t finish” — that distinction is what TITAN-AIWELCOME-04 depends on, and it doesn’t hold for every possible implementation (a system that times out silently may not know which case it’s in). Dated 2026-09-02; revisit on the first real implementation of any search, summarize, or recommend panel, and whenever EmptyState’s own known issues (its fallback copy, its heading-level defects) are resolved, since this page’s no-results variant composes that component directly.

AI Experience overview — vocabulary, agency tiers, and the six disclosure questions this page answers
Disambiguate — too many candidate answers, rather than none
Recommend — whose own empty case hands off to this pattern’s no-results treatment
Wayfinders: Initial CTA — a related but distinct first-encounter moment; see Gaps for the open boundary between them
Empty and zero states and EmptyState — the deterministic version of this pattern, composed directly by the no-results variant
Filtering and search — the manual fallback a user reaches for to check a no-results claim
Input, Tag, Button — the components this pattern composes

Why it works this way

Setting scope at the first interaction is cheaper than absorbing a refusal later. A user who learns upfront that the feature summarizes and surfaces Signals but doesn’t touch routing will never ask it to change routing and get an unexplained no; the cost of stating a limitation once, plainly, is far lower than the cost of a confused first request. Naming the specific, completed search is what keeps “nothing found” from reading as “something broke.” The two states look almost identical — an empty region with a short message — and the only thing separating a trustworthy empty result from a suspicious one is whether the message proves a real search happened. A heading that repeats the user’s own query back to them is that proof; a generic “no results” is not.

Overview

Wayfinders

Inputs

Tuners

Governors

Trust builders

Identifiers

Actions

Welcome / empty state

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