Who needs the state of AI-assisted data visualization?

The short answer

The audience is not “people interested in AI.”

Interest is cheap and difficult to interpret. The strongest audience consists of people who must decide where AI belongs in visualization work, what evidence to require, and who remains responsible when the output leaves the prompt box.

01 / Lead with decisions

A report is rarely the job.

Readers arrive because a pilot, deadline, research design, newsroom policy, or learning question has created a fork. Give them an artifact that helps them cross it.

02 / Keep one evidence base

Different routes should not create different truths.

The practitioner checklist, BI scorecard, research crosswalk, and newsroom protocol should reuse the same definitions, claims, limitations, and evidence dates.

03 / Earn distribution

Communities are relationships, not inventory.

Bring a native contribution, follow the host's rules, and ask for a review or discussion that fits the venue. A report link by itself is not participation.

04 / Measure changed work

Visibility is the first gate, not the outcome.

An indexed page, referral, citation, completed scorecard, changed pilot, and prevented error are different forms of evidence. Keep them separate.

What AI chart and dashboard systems can do in 2026, where they fail, and how to evaluate them for real work.

Recommended public positioning

One acquisition answer, seven translations

Twelve install signals. Zero retained-use rows.

Practice uses popularity only to shortlist a local trial. BI asks for successful use, governed acceptance, incidents, and whole cost. Research separates listing, installation, invocation, retention, organization, and outcome denominators. Newsrooms keep editorial and reader authority outside repository attention. Builders instrument attempt, success, invocation, return use, removal, and version with privacy and opt-out. Learning and access require representative use and transfer. Sponsors get the shortest defensible answer: twelve named listings collapse to ten parent repositories and eleven documented lineages; 0/12 exposes public invocation, retention, organizational acceptance, or outcome. Read the complete account →

One governance answer, seven translations

Three control contracts. Two use reports. Zero complete joins.

Practice asks which rules govern the exact task. BI asks whether all nine same-deployment receipts exist. Research separates configured controls from observed behavior. Newsrooms keep editorial and reader authority distinct. Builders instrument identity, routes, events, refusals, escalation, and version. Learning and access require representative return use. Sponsors get the shortest defensible answer: three provider families expose controls, two provider stories report named-feature use, two studies add adjacent governance practice, and 0/7 held rows joins authorization, effective scope, audit, routine use, incident disposition, independently accepted outcome, and later recheck. Read the complete account →

One evaluation answer, seven translations

One longitudinal development decision. One operational near-miss. Zero full episodes in 15 named cases.

Practice carries test cases, diffs, overrides, configuration, and artifact hash into acceptance. BI joins the evaluation export to release, authorized decisions, incidents, and recheck. Research keeps twelve lifecycle receipts separate. Newsrooms turn backend evidence into an editorial gate, not reader proof. Builders bind selection rationale to release and drift. Learning and access separate developer confidence from representative outcomes. Sponsors get the shortest defensible answer: Lexara adds six developers, two weeks, 38 experiments, 57 newly authored cases, ten models, and six prompts; no held case reaches accepted downstream delivery, consequential audience decision or calibrated trust, whole cost, and post-release return. Read all seven decision layers →

One primary-evidence answer, seven translations

Four empirical cases. Three receipt lanes. Zero deployments.

Practice keeps each human receipt with its artifact. BI requires an accepted build, routine use, outcome, later event, and affected-user return. Research preserves five full texts, 19 substantive surfaces, including one further full chapter, and leaves three DOI rows metadata-only. Newsrooms use literal stage verbs. Builders bind evaluation to release and recheck. Learning and access require representative use. Sponsors get the shortest defensible answer: four initial full-text empirical cases, seven further explicit human evaluations, one evaluation-to-redesign mechanism, and zero accepted deployments. Co-design, a demo, a human task study, and production intent are different receipts—not deployment. Read all seven decision layers →

One prototype-evaluation answer, seven translations

Seven human evaluations are still not one deployed lifecycle.

Practice keeps participant, task, response, and artifact state together. BI separates InfoViP's seven FDA evaluators and addressed feedback from its planned production installation. Research preserves the 22/9/8/7/14 coverage states. Newsrooms use controlled study, prototype evaluation, formative pilot, production intent, accepted release, routine use, and recheck literally. Builders bind the accepted build to telemetry, outcome, later change, and return. Learning and access require representative use on real access paths. Sponsors get the shortest defensible answer: eight substantive surfaces, seven human evaluations, one AI-adjacent non-GenAI production-intent near-miss, and zero complete production lifecycles. Read all seven decision layers →

One residual-recovery answer, seven translations

At that pass, the content gap was three. No new AI deployment lifecycle.

Practice gets an evaluation-to-redesign receipt: version, three UX experts, 25 problems, changed version. BI follows two practice-context near misses into accepted-build, ordinary-use, and outcome gates. Research preserves the historical 14/11/1/3 ledger. Newsrooms say expert-evaluated, applied with stakeholders, or evaluated across units. Builders freeze the changed build and return after change. Learning and access separate experts from affected audiences. Sponsors get the shortest defensible historical answer: all 14 residual rows investigated; eleven substantive, one full chapter, three metadata-only at that pass, zero complete AI lifecycles. Read all seven decision layers →

One actor-and-authority answer, seven translations

Three research contacts. Zero operating authorities. Closed is not shipped.

Practice keeps the case-series human-control path bounded. BI routes research-scope questions to Joshua Xu, Leihong Wu, and Oanh Dang while keeping operator, maintainer, QA executor, release approver, and authorization authority separate. Research codes the closed opportunity as an administrative state, not execution. Newsrooms say closed, not completed or shipped. Builders retain the nonemployee and governmental-authority boundary. Learning and access still require representative use. Sponsors require exact object, event date, role, application join, completed audit, and later return. The final component receipt remains real; the three research contacts do not fill the five operating-authority fields. Read all seven decision layers →

One review answer, seven translations

Five likely papers still do not complete the review frame.

Practice gets a 122 keyed / 5 unresolved receipt. BI keeps source label, review key, candidate, identifier validation, and admission authority separate. Research resolves five review-side joins before screening. Newsrooms may describe likely candidates only while saying A208 does not link them and must quarantine the unrelated PMID. Builders must abstain on five tempting matches and validate identifier ownership. Learning and access distinguish identity confidence from authority. Sponsors get the shortest defensible answer: 122 of 127 positions have review-owned keys; screening has not started. Never promote identity fit into provenance, reuse a conflicted PMID, or report zero of 127. Read all seven decision layers →

One production answer, seven translations

Three afterlives. Two restorations. Zero independent recoveries or authority transfers.

Practice asks whether the corrected route works for the person who failed. BI assigns semantic, deploy, incident, release, and acceptance owners. Research preserves actor and state on every transition. Newsrooms require rebuild, republish, editor check, and reader-surface recheck. Builders instrument proposal, merge, deploy, maintainer check, independent check, and prevention. Learning and access switch actors and devices. Sponsors get the shortest defensible answer: 3/3 cross delivery into a later event, 2/3 reach maintainer-verified restoration, 0/3 reach affected-actor recovery or transferred authority, and 0/3 clear all twelve states. Read the complete account →

Choose your route

Start with the decision in front of you.

Select a reader path to see the most useful first artifact, the question it answers, and the sections to read next. Nothing is hidden; the selector only gives this long report a sensible entrance.

Primary audience · route 01

Visualization and data practitioners

Where can AI remove work without taking away analytical intent, precise control, or accountability?

Trigger: A new tool, a tight deadline, an unfamiliar stack, or a generated chart that is almost—but not quite—right.
First artifact: AI visualization workflow evaluator
Decision: Which stages to delegate, assist, or keep manual; what to inspect; and when to stop or revert.
Read next: Practitioner dossier →

The audience map

Four primary audiences. Two specialist routes. One concise executive door.

Priority reflects fit with the current evidence, reachability through observable routes, and the presence of a concrete decision. It is not an estimate of market size.

Audience Decision moment Best first asset Likely route

P1 Visualization practitioners

Where AI belongs in a real workflow

Workflow evaluator

DVS · Nightingale · practitioner groups

P1 BI and analytics leaders

What to enable, buy, govern, or stop

Pilot and procurement scorecard

Tableau · Power BI/Fabric · dbt

P1 Researchers

Which study, benchmark, or claim survives

Benchmark crosswalk and gap ledger

IEEE VIS · labs · citable repository

P1 Journalists and editors

What is safe to use and publish

Newsroom acceptance protocol

NICAR/IRE · OpenNews · graphics teams

P2 Product and engineering teams

Which mechanisms and evaluations are durable

System anatomy and ablation guide

GitHub · Observable · engineering teams

P2 Educators and access specialists

What assistance preserves learning and access

Learning and accessibility protocol

Can we reproduce it?
Can an editor inspect the transformation?
What did the agent invent?
Which source, editorial, accessibility, and reader receipts are still missing?
What needs disclosure?
Can source material leave our environment?
Will a reader understand this on a phone?

What to give them

A newsroom-use protocol, provenance checklist, short benchmarking guide, and worked example that begins with source custody and ends with the delivered reader surface.

Trigger A roadmap choice, evaluation failure, customer demand, or plan to add an agent or skill.

Decision Context architecture; model versus deterministic component; equal-resource ablation; lane-specific eval suite; actor-separated recovery path; topology-bound phase instrumentation.

Language Construct validity, model-harness vintage, MCP, tool calling, agent skills, structured representations, critics, browser evaluation, phase trace, effective run manifest, repair loops, observability.

Trigger Curriculum redesign, assignment policy, accessibility review, or evidence of assisted completion.

Decision Allowed assistance; skills to teach; unassisted transfer; whether task success actually reaches participation, alternative pathways, assistive use, and delivered-reader acceptance.

Language Visualization literacy, transfer, co-creation, multimodality, verification, user agency, screen reader, low vision, and sustained use.

Jobs and decisions

The report becomes useful when it leaves a decision record.

The adjacent “automated consulting” work contributes one crucial correction: one corpus can support many renderings, but each rendering still needs a named user, trigger, decision, output, and acceptance test.

Language recommendation

~~Automated consulting~~ Decision products internally; evidence-backed decision guides publicly.

Name the object and its use. Do not make the production mechanism or labor replacement the headline.

Decision productTrigger and decisionOutputAcceptance

Practitioner

AI visualization workflow evaluator

Starting or repairing work → delegate, assist, or keep manual at each stage.

A stage-by-stage workflow and review plan.

The user can choose a path and name retained human responsibilities.

BI leadership

Pilot and procurement scorecard

New license, pilot, renewal, or incident → pilot, buy, constrain, or stop.

Readiness score, blocking conditions, pilot design, and fixed-to-downstream cost ledger.

A named owner accepts scope, volume, useful life, and amortization; wrong-number and recovery tests pass; planned and observed resources reconcile by phase and effective topology.

Research

Benchmark and gap crosswalk

Study design → reuse, extend, or reject an evaluation.

Comparable study cards and an unanswered question.

Every comparison is traceable; direct findings, cross-literature bridges, and later evidence lanes remain distinct; the reader can state what result would change the synthesis.

Newsroom

Use and publication protocol

AI-assisted analysis proposed → permit, revise, disclose, or prohibit.

Reproducibility and publication checklist.

A second person reproduces the result; editor and representative-reader checks pass.

Product team

Agent evaluation protocol

Feature, skill, model, or architecture choice → ship, change mechanism, or test further.

Equal-budget test and defect ledger.

Results survive equal-resource, deterministic, and rendered checks on the current model and harness; scope, regressions, and missing human lanes stay visible.

Education and access

Learning and access protocol

Curriculum or interface review → permit assistance, revise teaching, or reject delivery.

Assisted and unassisted outcome record.

Retention, transfer, or representative accessibility evidence—not completion alone.

Reusable specification

Every derived asset carries the same eight fields.

01
For whom
Role, environment, existing practice, and authority.
02
At what moment
The observable trigger that makes the material timely.
03
Decision
The actual fork, including a legitimate “do not use AI” branch.
04
Inputs
The local context the reader must supply.
05
Evidence
The report claims and limitations supporting the guidance.
06
Output
The record the reader leaves with, not merely information read.
07
Acceptance
How to know the decision path worked—and whether the evidence is accepted change, repeat contribution, maintenance authority, or independent recovery.
08
Reopen condition
What new product, model, benchmark, or field evidence forces review.

The language map

One canonical term. Several honest vocabulary bridges.

Use AI-assisted data visualization as the umbrella. It covers human-led work without claiming autonomy. More specific terms belong in route titles, questions, and metadata where they reflect the reader's real context.

Broad discovery

“AI data visualization tools”

AI chart generator · AI dashboard generator · ChatGPT data visualization · best AI for data analysis

Use in plain-language route copy and FAQs. Do not promise an unsupported “best tools” ranking.

Enterprise BI

“Conversational analytics”

Power BI Copilot · Tableau Agent · generative BI · agentic analytics · semantic model · certified metrics · governance

Use in the BI route. Keep provider feature terms distinct from independent evidence.

Practitioner workflow

“Can I edit the result?”

First draft · live data · precise edits · accurate numbers · show its work · DAX · SQL · handoff · maintenance

Lived friction is often more legible than “capability frontier.”

Research

“Agentic visualization”

LLM-based visualization generation · text-to-vis · chart-to-code · visual analytics · mixed-initiative · human-AI collaboration

Reserve “agentic” for systems with meaningful planning or action, not every AI-assisted step.

Journalism

“Coding agents for data analysis”

Practical AI in the newsroom · build your own AI benchmark · reproducible workflow · disclosure · reader trust

Lead with editorial work and accountability, not the detached phrase “AI visualization.”

Education and access

“Access starts before remediation.”

Co-creation · multimodality · verification · user agency · alternative pathways · maintenance · transfer · low vision

AccessViz treats AI-mediated access as a trust and power question. Outcome language belongs in the core route.

Evidence ladder

Language evidence is not demand.

Captured wording can improve an audience route. It cannot tell us that readers want this package, use the route, or make a better decision. Name the rung every time.

01 · observed now

Source wording

A term or question appears in captured community, editorial, implementation, or research material.

Earns candidate route copy

02 · after publication

Qualified discovery

A site-specific query or referral reaches the relevant route, with anonymization and truncation limits visible.

Does not prove use

03 · reader session

Decision-path use

A reader chooses a route and completes its scorecard, protocol, or crosswalk.

Does not prove outcome

04 · bounded outcome

Decision consequence

A pilot, workflow, study, publication, curriculum, or access decision changes, is confirmed, or stops.

Does not prove population impact

Independent editorial

Actual work, not intended workflow

PolicyViz and Visualising Data ask how AI fits real tasks, where it works, what a chart is for, and whether text interfaces remove useful creative friction.

One interview and one curated issue; not prevalence.

Spanish public-data practice

Question → context → inspectable result

Natural-language questions become analysis through dataset context, executable code, retries, corrections, and visible results.

One implementation account; not a language-market claim.

Accessibility community

Co-create, verify, challenge, sustain

AccessViz makes participation, agency, alternative pathways, maintenance, and delivered usability part of AI evaluation.

Workshop synthesis; participant denominator unstated.

Search measurement

Use the site's queries, with limits

Search Console can later show some qualified discovery. Trends is sampled, normalized context—not this site's demand.

No site-specific query evidence entered this update.

Portuguese teaching practice

Purpose → co-design → feedback → delivery

A 2026 USP course puts audience, user journeys, requirements, sketching, usability tests, feedback, and delivery alongside generative AI and Power BI.

One official 21-hour course record with 55 places; not a regional-market claim.

Fabric community · source wording

Declare scope, then retain correction

One practitioner issue makes capacity, Viewer/Build permission, and model traversal explicit; one Microsoft-authored case links wrong answers to context, inclusion, naming, filters, and validation.

Two C-grade self-reports; not a bug, breach, efficacy result, prevalence estimate, or demand.

Retained-human-value check

What should AI not optimize away?

Efficiency is an incomplete verdict. Each audience layer should preserve the work through which people set purpose, learn together, correct errors, and carry a result into use.

01

Purpose + audience

State why the visualization exists, who must understand or act, and what context the system cannot infer.

02

Collaboration + creative friction

Keep co-design, critique, exploratory detours, teaching, and the interpersonal work through which knowledge is made.

03

Inspectable correction

Trace extraction and encoding choices, repair errors, and check whether a repair introduced another problem.

04

Feedback + delivery

Carry a draft through user feedback, usability and accessibility testing, handoff, deployment, maintenance, and actual use.

Discovery doors

Questions worth answering verbatim

Each deserves a direct two- or three-sentence answer, a visible evidence date, and a route into the deeper report.

What can AI actually do in a data-visualization workflow in 2026?
Can AI-generated charts and dashboards be trusted?
How should a team evaluate Power BI Copilot, Tableau Agent, or conversational analytics?
Which parts of data visualization should remain human-controlled?
Does an AI visualization tool save time after checking and repair?
What is agentic visualization, and how is it evaluated?
Which benchmarks test chart generation, dashboards, interaction, or reader understanding?
How should journalists disclose and validate AI-assisted analysis and graphics?
What visualization skills still matter when AI can produce a first draft?
How do accessibility and mobile testing change the verdict?

Public artifact architecture

One canonical hub. Many ways in. No sibling islands.

The current long-form artifacts already have crawlable text, literal titles, responsive layouts, section anchors, and visible evidence dates. The missing layer is the shared public entrance, citation/version system, audience router, and reusable evidence bundle.

Canonical entrance

Evidence hub + audience router

One durable URL, thesis, evidence date, scope, findings, routes, and links to every rendering.

Read quickly

Executive brief

Five pages and two figures.

Understand fully

Comprehensive report

The connective argument and evidence accounting.

Enter by question

Standalone essays

Self-contained findings with local context.

Make a decision

Guides and scorecards

Inputs, output, acceptance, and reopen condition.

Reuse the evidence

Tables, figures, and data

Public-safe exports with stable anchors and sources.

Audit the work

Methods and versions

Limits, corrections, citation, frozen releases, and change log.

Route page contract

Within one screen, tell the reader whether this can help.

MomentYou are deciding whether to…

Short answerThe strongest supported conclusion.

Decision toolA checklist, scorecard, or crosswalk.

Three findingsEach linked to evidence and limitation.

Worked caseA realistic path through the decision.

Go deeperRelevant report sections and methods.

RecheckEvidence date and next update trigger.

Search and AI visibility

Build a legible publication, not an AEO ritual.

Google's current guidance points to the same foundations for ordinary Search and its generative AI features: useful original content, crawlable text, clear structure, semantic HTML, indexability, and good page experience. It says no special AI markup, tiny “AI-sized” chunks, or llms.txt file is required for Google.

Stable descriptive URLs and canonical links
Sitemap and feed for material updates
Unique titles, descriptions, authorship, and dates
Stable figure and table anchors
Public-safe CSV or JSON where the material is genuinely data
Preferred citation and frozen release after approval
HTML as the primary reading surface; PDF as a companion

Official Google guidance →

The rollout

Reader proof before reach. Interpretation before repetition.

The report launches once. Its audience routes travel over several weeks. The order is designed to replace modeled usefulness with observed decisions as quickly as possible.

Phase0

Days 1–14

Prove decision usefulness

Recruit 8–12 named reviewers, with at least two from each primary audience. Give each one scenario rather than the whole site and a satisfaction survey.

Can they identify their route?
Can they find evidence for a named decision?
Can they distinguish product claims, benchmarks, human evidence, and unknowns?
Can they produce the intended decision record?
Can they name what remains unresolved?

Phase1

Days 15–30

Publish the canonical package

Release the hub, audience router, executive brief, comprehensive report, long-form companions, initial decision tools, methods, public-safe evidence, correction policy, change log, and reusable figures together.

One launch claim
One original figure
One clear audience promise
One route into the full package

Phase2

Days 31–75

Distribute through native audience routes

Carry one useful artifact at a time through the communities that already own the context.

Visualization practice: an original Nightingale essay or DVS discussion centered on the workflow evaluator.
BI and analytics: a concrete scorecard session for a product or analytics-engineering group.
Research: the citable crosswalk and gap ledger to labs and workshop participants.
Journalism and education: separate newsroom and learning/access protocols.

Phase3

Days 76–90 and onward

Keep the evidence alive

Publish change notes only when material claims, sources, tool contracts, benchmarks, or forecasts change. Preserve dated releases and turn repeated reader questions into better routes or tools.

Review volatile product statements internally.
Publish a quarterly evidence delta when it contains real change.
Invite corrections and missing primary sources.
Publish experiments when they add evidence, not to fill a quota.

Channel map

Bring the contribution each surface is built to receive.

RouteNative contributionParticipation ruleFirst ask

Nightingale

Original essay, case, or method

Join and submit under editorial and exclusivity terms

Editorial fit for one standalone finding

DVS groups and events

Discussion, peer review, event, or approved research recruitment

Follow channel rules; no repetitive link dropping

Review one workflow evaluator or field question

Tableau and BI groups

Live governance case or scorecard session

Use a product-specific scenario and disclose independence

Test pilot readiness against a real case

dbt community

Semantic-layer or governance discussion

Tie the visualization claim to analytics-engineering work

Review context and readiness inputs

IEEE VIS and labs

Paper, dataset, position, or reading-group discussion

Respect closed deadlines and archival standards

Correct the benchmark crosswalk and gaps

NICAR and OpenNews

Training, session, protocol, or operational case

Serve newsroom work rather than promote a product

Test a reproducible acceptance protocol

GitHub and archive

Versioned report, public-safe data, methods, corrections

Clear license and custody boundary

Reuse or challenge a stable artifact

Newsletters and podcasts

One figure, case, interview, or defensible debate

Pitch a story for that audience, not a generic blast

Discuss one finding and its consequence

Evidence of usefulness

Do not let page views swallow the outcome.

A useful measurement system preserves the full chain from technical eligibility to changed work. Each level answers a different question and proves less than the level after it.

01

Eligibility

Can the right reader or system retrieve it?

Crawl · index · metadata · HTTP · mobile

Does not prove notice

02

Qualified circulation

Did intended readers reach the relevant route?

Referral · engaged reading · return use · method views

Does not prove a changed decision

03

Decision use

Did someone use the artifact for its intended job?

Scorecard · protocol · decision record · internal citation

Does not prove a good outcome

04

Reuse

Did the evidence travel with attribution?

Backlink · academic citation · syllabus · figure · method

Does not prove correct interpretation

05

Relationship

Did qualified readers consent to continued contact?

Update opt-in · correction · interview · invitation

Does not prove adoption

06

Outcome

Did work actually change?

Pilot changed or stopped · policy revised · study adopted · error prevented

The evidence we ultimately want

First-cycle learning thresholds

Small enough to observe. Strong enough to matter.

8–12 prelaunch decision-path reviews across four primary audiences.
One corrected route or artifact from each audience's feedback.
Three completed uses of each launch decision tool.
Five attributable reuses in protocols, studies, syllabi, presentations, or technical work within a quarter.
Two outcome accounts, including “we narrowed or stopped the pilot” as valid positive evidence.
One visible ledger of substantive corrections and missing evidence.

Failure modes

What this rollout is designed to prevent.

The long report is the only product: Route by decision and create outputs readers can use.
A tool-ranking frame wins: Explain families and evaluation; refuse unsupported “best” claims.
Community posts feel promotional: Earn access and bring a native contribution.
Private source custody leaks: Publish only public-safe synthesis, metadata, and allowed excerpts.
Updates destroy citability: Freeze versions and maintain a visible change log.
Cadence creates thin content: Publish on evidence changes and repeated reader questions, not quota.

Ownership and gates

Research readiness is not publication authority.

The work crosses private source custody, synthesis, public presentation, human relationships, and outcome research. Keeping those owners separate is part of the evidence design.

Private evidence

Evidence owner

Captures, grades, claim links, research questions, and provenance. Private source contents do not become public because the synthesis is ready.

Synthesis

Research owner

Cross-source relevance, recommendations, experiment agenda, and this audience strategy. Every material public claim retains its evidence class and limit.

Presentation

Public artifact owner

Report design, routes, stable URLs, accessibility, content parity, metadata, and public-safe reusable evidence.

Authority

Human owner

Publication, authorship, licensing, DOI, submissions, outreach, and participation. Approval for one surface does not authorize every channel.

Relationships

Named participant

Community contribution, editorial conversation, invitation, and follow-through. No synthetic personas or unsupervised promotion.

Outcomes

Audience research owner

Decision-path evidence, reuse, relationship, and changed work—kept separate from traffic and impressions.

Evidence and limits

The audience map is a testable model.

It combines direct inspection of the current report package, official descriptions of community and publication routes, conference programs, search-measurement and citation guidance, and bounded public language samples. It contains no site-specific query or observed decision-path evidence.

Observed

Current local reports and rendered artifacts; official access and editorial terms; official event programs; official search and citation mechanics; bounded public language.

Inferred

Audience priority, decision moments, useful artifact forms, the sequencing of routes, and the 90-day rollout.

Unknown until tested

Who reads, refers, cites, subscribes, invites, completes a decision tool, or changes work because of this package.