> ## Content Index > Fetch the complete content index at: https://www.agenticlandmark.com/llms.txt > Use this file to discover other available public pages before exploring further. # Reach is not Proof - URL: https://www.agenticlandmark.com/brief/reach-is-not-proof/ - Published: 2026-10-31T00:41:37.000Z - Updated: 2026-10-31T00:41:37.000Z - Description: Sort shopping agents by evidence instead of audience and the order changes. Three have published purchase results, all run by retailers on their own data. Two have nothing measured yet. - Author: Jim Cook - Tags: agentic-commerce, data, amazon, walmart, openai, adobe, retail *Every shopping agent can claim a share of your customer's attention. Only a few can show what that attention bought.* The question a CMO asks after mapping the five kinds of shopping agents is which one to prepare for first. The usual answer is the biggest. That sorts agents by reach, and reach is the easiest thing in this market to measure. Sort them by evidence instead, by what each has shown it does to a purchase, and the order changes. The agents with the largest audiences are not the agents with the clearest commercial results, and two of the five have no published results at all. What follows is that scoreboard, as of October 2026: which agents have proof, which have reach, which have nothing measured yet, and what the picture looks like product category by product category. ## The agents with proof Three agents have published results tied to purchases, and all three belong to retailers. Amazon says its shopping assistant helped deliver nearly $12 billion in incremental annualized sales in 2025, and that shoppers who used it were 60 percent more likely to complete a purchase. The assistant, Rufus until May 2026 and now Alexa for Shopping, has been used by more than 350 million customers in the twelve months to June 2026\. Those are Amazon's own published figures. Walmart's CEO, John Furner, said on the company's August 2026 earnings call that customers using Sparky spend 40 percent more per order than those who do not, and that the number of Sparky customers is up 70 percent year over year. Lowe's reports that conversion more than doubles when customers engage with its Mylow assistant online. The common thread is not size or model quality. Each agent runs on data its operator owns: the full catalog, live inventory, purchase history, loyalty status. [Walmart built its agent. Amazon will rent you one.](https://www.agenticlandmark.com/brief/walmart-built-its-agent-amazon-will-rent-you-one-the-difference-is-the-data-layer/) Made that case for Walmart in May. The scoreboard extends it beyond Walmart: so far, every agent with published purchase results runs on its operator's own data. ## The agents with reach The general-purpose platforms have the audiences. OpenAI confirmed 900 million weekly active users of ChatGPT in February 2026\. Google says people shop across Google more than a billion times a day, against a Shopping Graph of more than 60 billion product listings. What they have not shown is a purchase result of their own. OpenAI scaled back in-chat Instant Checkout in early 2026, and Walmart measured in-chat checkout converting roughly three times worse than a click through to walmart.com. On the advertising side, Criteo's CFO said on the company's first-quarter 2026 call that its ChatGPT Ads beta, with more than a thousand advertisers onboarded, contributes no material revenue yet. That does not make reach worthless. It means the result shows up somewhere else: on the brand's own site, as traffic an agent sent. Adobe's July 2026 data has AI-referred retail visits converting 60 percent better than other visits. In April 2026, Dell's head of global consumer revenue programs told Digital Commerce 360 that agentic shopping was delivering nothing close to earth-shaking for Dell. Both are true. Adobe measures the visits that arrive across many retailers. Dell measured one site. A strong aggregate is consistent with individual brands seeing nothing, and the likeliest difference is whether an agent could read the brand's products well enough to recommend them. ## The agents with nothing measured yet Two of the five kinds have no published purchase results. Vertical agents, built for one category such as fashion or travel, have produced none the practice can verify. White-label agents are only now adding checkout: Constructor launched in-agent checkout for its retail shopping agent on October 5, 2026, with no volume reported. An empty row is not a failing grade. It means a brand cannot yet size these agents from evidence, only from a vendor's pitch. That is a reason to treat any forecast about them as a forecast, and to put readiness money first where something has been measured. ## Category by category The same sort, done by what a brand sells rather than which agent it faces, shows how uneven the picture is. **Travel** is moving fastest, and the movers are aggregators, not hotel or airline brands selling through a cart. Expedia is building an agent inside Hotels.com and agentic capability for Amazon's Alexa+, and named generative and agentic AI as a competitive threat in its annual report. Trade coverage of Adobe's July 2026 data puts AI-referred travel traffic at roughly parity with other sources on conversion, a gap that was 28 percent in May. **Beauty and apparel** have the best-known names. Glossier, SKIMS, Spanx and Vuori were among the brands announced as ChatGPT added shopping in February 2026\. That is evidence of being discoverable and evaluated, not of sustained sales through the agent. **Electronics** had no confirmed recognizable brand transacting on an agent surface as of July 2026\. The signal is announced pipeline, not live volume. **B2B** activity is real but led by procurement platforms and source-to-pay suites, not by end brands. Orders carry negotiated pricing, credit terms, and ERP integration, so near-term B2B buying is assisted by agents rather than run by them. Across every category, the defensible claim is the same: discovery and evaluation are already agent-mediated. Claims about who is selling through an agent age badly. Claims about who is being evaluated by one hold. ## What the scoreboard tells a brand Put the rows together and the uncomfortable part is plain. The agents with proof belong to the retailers a brand sells through, and the brand does not control their data. The agents with reach read only what the brand publishes, and have not yet shown what that reading is worth. Neither half is a reason to wait. The proof shows that agents running on complete product data move purchases. The reach shows where the next audience already is. What connects them is the one input a brand controls in every case: whether its own product data is complete enough for any agent to compare. That is where readiness spending earns its keep while the scoreboard fills in. [Five kinds of shopping agents are already evaluating your products](https://www.agenticlandmark.com/brief/readiness-for-one-shopping-agent-does-not-transfer/); maps what each kind of agent needs. The [Agent Catalog Read](https://www.agenticlandmark.com/agent-catalog-read/) shows what an agent can extract from your product pages, product by product. Reach decides whether you are considered. What an agent can read decides whether you are chosen.