AI Search

Measure AI shopping by task

Separate discovery, purchasing and service questions using Salesforce’s historical shopper survey, complete response tables and a practical AI visibility review workflow.

Three separate paper cards show a magnifier, shopping basket and support speech bubbles

Key takeaways

  • Salesforce’s late-2024 survey reports AI use separately for purchasing, service and discovery. Keep those task labels in your reporting. [1]
  • Ansvra’s sums of the three positive frequency categories are 30% for purchasing and 39% each for service and discovery. These are separate task distributions. [1]
  • Retain all five response categories, including Never and Don’t know; rounded source rows can total 99%. [1]
  • Match each visibility observation to a concrete customer question, and obtain separate evidence before claiming a completed purchase or resolved service need.

Label AI shopping questions by the task they address before combining their results. A recommendation, an answer about returns and a maintenance explanation can all matter to the same brand, while supporting different decisions. A saved answer tells you what appeared for a question. Establishing that someone discovered a product, completed checkout or solved a problem requires additional evidence.

Salesforce’s sixth Connected Shoppers Report provides a useful historical example. Its fieldwork ran from November 27 to December 26, 2024; the edition was announced on March 24, 2025. The selected chart asks about three AI shopping tasks and shows a different response distribution for each. Treat these as dated survey findings, rather than current adoption estimates. [1] [2]

What the three shopping tasks show

The chart below recreates all five frequency categories for all three task rows on page 22 of the report. Discovery and service each have a larger combined positive-category total than purchasing, but their frequency mixes differ. You can inspect each category, toggle series and download the chart’s complete scoped data.

Reported AI shopping use by task

Salesforce’s shopper survey, November 27–December 26, 2024. All five frequency responses are retained. Category key: Freq. = Frequently; Occ. = Occasionally; 1–2 = Once or twice; D/K = Don’t know. Never is unchanged.

1007550250PurchasesServiceDiscoveryTask
  • Freq.
  • Occ.
  • 1–2
  • Never
  • D/K
Full data table
Reported AI shopping use by task · % of surveyed shoppers
TaskFreq.Occ.1–2NeverD/K
Purchases11712664
Service11919555
Discovery121017564

Each row describes a separate task. Values are rounded, with displayed totals of 100%, 99% and 99%. Ansvra recreated this chart from the public report and kept the original values. These are independent task marginals, not a sequence of shopper outcomes. Labels: Purchases = Making purchases; Service = Receiving customer service; Discovery = Product discovery and inspiration. Source: Salesforce, Connected Shoppers Report, sixth edition, page 22. © 2025 Salesforce, Inc. [1] [2]

Source: Salesforce, Connected Shoppers Report, sixth edition, page 22. Fieldwork: November 27–December 26, 2024. The chart concerns shoppers’ reported use of AI broadly. It does not identify a particular assistant, search experience or autonomous purchasing product. The supporting table above retains every printed value used in this analysis. [1]

For a compact comparison, Ansvra adds Frequently, Occasionally and Once or twice within each task. The following table shows the complete calculation scope alongside a check of all five displayed categories.

TaskPositive-category sumAll five categories
Making purchases30%100%
Receiving customer service39%99%
Product discovery and inspiration39%99%

These sums come from rounded published cells. The discovery total also agrees with the 39% figure in Salesforce’s announcement. Discovery minus purchasing is 39 − 30 = 9 percentage points. That difference describes these displayed task totals; it does not measure a change over time or establish a statistically significant difference. [1] [2]

Why this cannot become a conversion funnel

A tempting reporting shortcut is to put discovery at the top, purchasing beneath it and service after it. That might look like a customer journey, but this chart does not provide the connections between those stages. Each row summarizes answers about a separate task. The publication does not give the respondent-level combinations needed to connect the rows.

Dividing the purchasing total by the discovery total would therefore produce an unsupported conversion claim. We do not know which purchasing respondents also reported discovery use, whether either task came first, or whether they involved the same product. Even a person who selected a positive response for both tasks would not, by that fact alone, establish that AI discovery caused their purchase.

The equal service and discovery sums deserve similar care. Their underlying cells differ: service has a larger Once or twice value, while discovery has larger Frequently and Occasionally values. Equality after adding categories does not establish equal habits, identical users or equal commercial importance. Keep the original categories available whenever you show the simplified totals. [1]

This matters for brand reporting because visible answer elements are also different kinds of evidence. A cited buying guide can support an observation about a reference in an answer. A product card can support an observation about a shopping placement. Neither observation supplies the missing link to completed orders, customer satisfaction or the business value of a service interaction.

Keep the method next to the numbers

Salesforce describes two double-anonymous surveys of third-party panelists: 8,350 shoppers and a separate group of 1,700 retail industry decision-makers. The report describes respondents across 21 countries and five continents. Only the shopper task-frequency chart enters this article’s calculations; the retailer sample is not part of its denominator. [1]

The overall shopper sample size gives useful context, but the selected chart does not publish item-specific response counts or numerator counts. Multiplying a rounded percentage by 8,350 would create an apparent count that the chart does not report. We retain those missing counts as unknown and describe the results as findings among participating shoppers across the surveyed countries.

The public material also leaves several interpretive limits. It does not provide a full questionnaire, precise recall window, quantitative definitions for each frequency label, chart-specific weighting details or uncertainty intervals here. Frequently should therefore remain the source’s category, rather than being relabeled daily. Broad AI use should remain broad, without becoming a claim about a named platform or an agent completing a transaction.

Rounding is another practical trap. The five displayed values total 100% for purchasing and 99% for both service and discovery. Salesforce warns that rounded report totals can differ from 100%. Preserve these values rather than scaling them to fill a complete bar. Calculating any use as 100 minus Never would also include Don’t know and the rounding difference. Add the three specified positive categories instead. [1]

Salesforce produced the research and has a commercial interest in retail AI. This article uses its attributed shopper findings separately from its product promotion and retailer forecasts. The figures are historical context for a measurement decision, not evidence that today’s shoppers use AI at the same rates.

A magnifier inspects an answer card beside three folders marked with shopping-task pictograms
Group the question by task before deciding which answer evidence to inspect.

Turn the task into a question you can investigate

Start with a customer need your team can explain. The following fictional examples show how a retail team might distinguish questions before reviewing answers. They are planning examples, with no implied observations or performance results.

Discovery: assess suitability

A shopper might ask which lightweight rain jacket suits a short urban commute. Your content question could be whether the answer explains relevant fit, weather and carrying trade-offs, and which pages it cites. A brand mention alone would leave those suitability details unresolved. Use customer research to decide which needs deserve tracked questions, then inspect the available answer evidence.

A question about whether a jacket can be returned after trying it on concerns a purchase decision, but it is still an information request. You could check whether the answer accurately reflects the applicable policy and references an appropriate page. This example is narrower than the survey’s Making purchases label. Recording it does not establish a checkout or justify counting a sale.

Service: support use after purchase

A question about cleaning the jacket without damaging its finish calls for care instructions. A useful review would check the advice against the product’s actual guidance and inspect available references. An answer that cites a care page is evidence of that citation. Confirming that a customer followed the advice and resolved their problem needs a different measurement.

These examples give content teams distinct follow-up work: improve suitability evidence, clarify a policy, or make product guidance easier to use. Choose the action from the answer’s actual omissions and the customer question, rather than distributing effort according to three historical survey percentages.

A review workflow for task-specific visibility

The diagram describes an editorial review process: define the task, choose the question, open the answer, inspect the evidence and state the next check. It is an original suggested workflow, not a reconstruction of shoppers moving through the survey’s rows. Select a step for its explanation.

Match a shopping question to the evidence you report

Ansvra’s suggested five-step review workflow for task-specific visibility reporting.

  1. Define task

    Identify discovery, purchasing or service as the question under review. The source chart reports each separately. [1]

    Next: Choose question

  2. Choose question

    Use customer evidence to select a concrete question worth tracking. The survey’s marginals are context for this choice, not budget weights. [1] [2]

    Next: Open answer

  3. Open answer

    Read the relevant saved response and record the date and question. This is evidence of that response rather than a shopper’s completed journey. [1]

    Next: Inspect evidence

  4. Inspect evidence

    Review available cited pages, answer content and shopping placements for that question. Record unavailable evidence clearly. [1]

    Next: Next check

  5. Next check

    State what was observed and which outcome would need separate evidence. A task-frequency distribution cannot establish conversion. [1]

This is an original editorial workflow, not an observed shopping funnel. Salesforce supports the survey scope and task definitions. [1] [2]

In Ansvra, AI Prompt Research helps teams research relevant questions and select what to track. Add your task label and reason for choosing the question to the working brief. A survey distribution can motivate looking at several tasks, while your own customer evidence determines which questions matter for your brand.

When reviewing a relevant saved response, AI Shopping & Ads helps inspect available answer content, product cards and sponsored placements. Keep an observed element, none observed and unavailable evidence distinct. This allows a report to describe what the saved response contains without extending that observation into a claim about checkout or service completion.

Use Citation Source Analysis to examine the cited pages and references in saved answers. A policy citation may suggest checking the currency of that policy; a guide citation may suggest reviewing its suitability explanation. The source reference gives your team somewhere concrete to investigate. It does not guarantee that a user visited the page or acted on its content.

A useful review note can stay short: identify the question and task, name the saved answer and date, describe the observed evidence, then state the next verification. For example, a team could plan to check a return condition against the current policy before deciding whether the page needs revision. Keep proposed checks distinct from completed checks and observed outcomes.

What to carry into your next report

Retain separate groups for discovery, purchase-related information and service when those groups represent different customer needs. Within each group, state which questions were reviewed and which answer evidence was available. Describe any change against a defined comparison rather than combining different question sets into an unexplained shopping score.

If a stakeholder asks how visibility affects sales or service resolution, identify the missing outcome evidence explicitly. The shopper survey cannot supply it, and an answer observation cannot substitute for it. That boundary gives the team a precise next research question while keeping the current report useful: it shows what people might ask, what the answer contained and which content decision follows.

Summary

Separate shopping questions by task, preserve the complete survey categories and keep the late-2024 study period visible. Review the answer and its available references or placements before deciding what content to improve. Use separate outcome evidence for claims about purchases or resolved service needs; task-frequency totals and AI visibility observations answer different questions.

Sources

  1. Connected Shoppers Report, sixth editionAccessed
  2. Salesforce research announcement: AI agents and retailAccessed

Discussion

Join the discussion

Start the conversation

Share a question or tell us what worked for you.

Comments appear after review. How we handle comments

The Ansvra journal

Keep reading

Browse all articles

Ansvra

A clearer next step.

MEET YOUR AI GUIDE

A little clarity.
A clearer next move.

Ask about Ansvra, AI visibility, or your next step with content.

Opening chat…