
How personal agents choose
Where Grok Bot, Muse, Instinct and Dots send shoppers, which products they pick, and what changes the pick.

Aaron Soerjohadi
GTM

Angela Felicia
Engineering

Kevin Goenadibrata
Engineering
- 11,200
- Personal-agent runs
- 1,255
- Prompts


- Grok Bot, Muse, Instinct, Dots
Destinations
Personal agents send 76.2% of their store and review citations to stores, against 61.6% for regular LLMs.
Store share of store and review citations
| Group | Assistants | Store citations | Store + review citations | Store share |
|---|---|---|---|---|
| Regular LLMs | ChatGPT, Gemini | 841 | 1,365 | 61.6% |
| Personal agents | Grok Bot, Muse | 2,008 | 2,634 | 76.2% |
Share of runs citing each site
| Site | ChatGPT | Gemini | Grok Bot | Muse | Instinct | Dots |
|---|---|---|---|---|---|---|
| Brand's direct store | 42.2% | 16.3% | 64.4% | 56.2% | 75.3% | 63.2% |
| Retail stores (Best Buy, Amazon or Walmart) | 21.7% | 13.0% | 37.0% | 38.6% | 13.9% | 13.1% |
| Best Buy | 7.3% | 2.4% | 11.9% | 21.5% | 4.6% | 5.9% |
| Amazon | 0.0% | 0.0% | 17.5% | 0.6% | 1.6% | 0.3% |
| Walmart | 15.7% | 11.0% | 14.0% | 26.4% | 7.8% | 6.9% |
| NBC News | 6.1% | 0.2% | 0.3% | 0.0% | 0.9% | 0.2% |
| 0.2% | 4.9% | 1.9% | 0.0% | 1.1% | 0.9% | |
| RTINGS | 6.5% | 0.4% | 3.7% | 1.9% | 2.1% | 11.4% |
Brand's direct store means a product, buy or pricing page on the maker's own site. A site counts once per run.
Share of answers linking to a place to buy
| Assistant | Type | Runs | Anywhere to buy | A store | Brand's direct store | Brand's store only, no retailer |
|---|---|---|---|---|---|---|
| ChatGPT | Regular LLM | 642 | 79% | 59% | 42% | 20% |
| Gemini | Regular LLM | 491 | 68% | 57% | 16% | 11% |
| Grok Bot | Personal agent | 579 | 99% | 82% | 64% | 17% |
| Muse | Personal agent | 637 | 99% | 88% | 56% | 11% |
| Instinct | Personal agent | 563 | 98% | 53% | 75% | 45% |
| Dots | Personal agent | 641 | 97% | 64% | 63% | 33% |
Share of answers citing each type of site
| Assistants | Retailers | Brand sites | Review sites | Press | Social |
|---|---|---|---|---|---|
| Regular LLMs (ChatGPT, Gemini) | 58% | 38% | 44% | 17% | 3% |
| Grok Bot + Muse | 85% | 62% | 41% | 15% | 5% |
| Instinct | 53% | 77% | 35% | 23% | 4% |
| Dots | 64% | 64% | 39% | 4% | 4% |
An answer can cite several types, so rows don't add to 100%.
Grok Bot + Muse: site types cited, by category
| Category | Retailers | Brand sites | Review sites | Press | Social |
|---|---|---|---|---|---|
| Groceries | 99% | 14% | 11% | 4% | 1% |
| Hotels | 99% | 15% | 3% | 1% | 0% |
| Beauty | 99% | 63% | 20% | 3% | 1% |
| Apparel and shoes | 98% | 58% | 41% | 10% | 4% |
| Computers for local AI | 89% | 84% | 60% | 30% | 20% |
| Electronics | 83% | 79% | 74% | 35% | 3% |
| Home appliances | 81% | 89% | 58% | 21% | 3% |
| Software | 1% | 99% | 66% | 7% | 12% |
| All categories | 85% | 62% | 41% | 15% | 5% |
Method
16 Agent Shopping Study products, ZIP-code variants left out. Site tables use the 77 prompts that ChatGPT, Gemini, Grok Bot and Muse all answered with sources: 2,361 runs (ChatGPT 642, Gemini 491, Grok Bot 587, Muse 641), plus Instinct 563 and Dots 641 on the same prompts. Site-type tables use 151 prompt and product pairs: 3,751 runs. The 470 most-cited domains (80% of agent citations) were typed by hand.
Products
Most head-to-heads have one clear winner. In 9 of 15, agents picked the runner-up in at most 1 of 12 tests. Second place barely gets picked.
- Under 60%
- 60%
- 75%
- 90%+
Consumer
| Matchup | Unbranded“What's the best…?” | Head-to-head“X or Y?” | Buy it“Pick one and take me there.” | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
![]() | ![]() | ![]() | ||||||||||
| Roborockvs iRobot | Roborock78% | Roborock78% | Roborock100% | Roborock90% | Roborock83% | Roborock100% | Roborock100% | Roborock100% | Roborock100% | Roborock100% | Roborock100% | Roborock100% |
| CeraVevs Neutrogena | CeraVe56% | CeraVe56% | CeraVe100% | CeraVe73% | CeraVe83% | CeraVe83% | CeraVe67% | CeraVe100% | CeraVe100% | CeraVe100% | CeraVe100% | CeraVe100% |
| Sonyvs Bose | Sony89% | Sony89% | Sony67% | Sony75% | Sony67% | Sony100% | Sony100% | Sony63% | Sony100% | Sony67% | Split 1–1 | Bose75% |
| Applevs Dell | Apple56% | No pick80% | Apple100% | Apple56% | Apple100% | No pick67% | Split 1–1 | Apple57% | Apple100% | Apple100% | Apple100% | Apple100% |
| Nikevs Adidas | HOKA56% | Brooks33% | Brooks50% | Saucony50% | No pick89% | No pick88% | Adidas67% | Nike73% | Nike100% | Nike100% | Nike100% | Adidas67% |
| Instacartvs Walmart | Instacart67% | No pick50% | Instacart67% | Instacart75% | No pick83% | No pick80% | Walmart100% | Walmart100% | Instacart67% | Split 1–1 | Walmart100% | No pick100% |
Software
| Matchup | Unbranded“What's the best…?” | Head-to-head“X or Y?” | Buy it“Pick one and take me there.” | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
![]() | ![]() | ![]() | ||||||||||
| Canvavs Adobe | Canva67% | Canva67% | Canva100% | Canva70% | Canva67% | No pick56% | Canva100% | Canva100% | Canva100% | Canva100% | Canva100% | Canva100% |
| DaVincivs Premiere | No pick67% | DaVinci75% | DaVinci100% | DaVinci50% | DaVinci100% | DaVinci75% | DaVinci100% | DaVinci100% | DaVinci100% | DaVinci100% | DaVinci100% | DaVinci100% |
| Slackvs Teams | Slack100% | Slack67% | Slack100% | Slack100% | No pick60% | No pick60% | Slack100% | Slack82% | Slack75% | Slack75% | Slack88% | Slack93% |
| Zoom / Google MeetNo clear winner | No pick78% | No pick56% | Zoom100% | Zoom83% | No pick56% | No pick75% | Google Meet67% | Google Meet75% | Google Meet67% | Zoom67% | Google Meet100% | Zoom75% |
| CapCutvs iMovie | CapCut100% | CapCut100% | iMovie67% | CapCut100% | No pick100% | No pick100% | iMovie100% | CapCut86% | CapCut100% | CapCut100% | CapCut100% | CapCut100% |
| Photoroomvs remove.bg | remove.bg67% | Split 1–1–1–1 | Adobe67% | Photoroom50% | Split 3–3 | Photoroom100% | Photoroom100% | Photoroom57% | remove.bg67% | Photoroom100% | Photoroom100% | Photoroom67% |
| Descriptvs Adobe Podcast | Descript67% | Descript67% | Descript60% | Descript67% | Split 3–3 | No pick100% | Adobe Podcast100% | Adobe Podcast100% | Descript100% | Descript67% | Descript100% | Descript100% |
| Acrobatvs Smallpdf | Acrobat100% | No pick67% | Acrobat100% | Acrobat100% | No pick100% | No pick83% | Split 1–1 | Acrobat57% | Smallpdf67% | Acrobat67% | Split 1–1 | Split 2–2 |
| Google Slidesvs PowerPoint | No pick78% | No pick44% | Split 1–1 | Google Slides80% | No pick100% | No pick83% | Google Slides100% | Google Slides100% | Google Slides100% | Google Slides100% | Google Slides100% | Google Slides100% |
Each cell is the answer that agent gave most often, with how many runs gave it. Hover a cell for the full label.
Method
Agent Head-to-Head prompts. Unbranded: “What's the best…?”. Head-to-head: “X or Y?”. Buy it: “Pick one and take me there.” Completed Grok Bot and Muse runs through Nov 5, 2026, Instinct through Nov 6, Dots through Nov 8.
Canva vs Adobe
Canva is the default, even on a tight budget. Name pro-level work and agents pick Adobe 41 of 41 times.
Pick by what the user asks for, all agents pooled
| Request | Answers | Canva | Adobe | No single choice |
|---|---|---|---|---|
| No specific requirement | 82 | 78% | 0% | 22% |
| Price or budget matters | 49 | 76% | 14% | 10% |
| Team needs shared templates | 14 | 57% | 43% | 0% |
| Beginner or daily social posts | 27 | 93% | 4% | 4% |
| Needs pro-level output | 41 | 0% | 100% | 0% |
Pick by agent
| Request | Grok Bot | Muse | Instinct | Dots |
|---|---|---|---|---|
| No specific requirement | Canva 65% | No pick 58% | Canva 100% | Canva 100% |
| Price or budget matters | Canva 79% | Canva 67% | Canva 79% | Canva 75% |
| Team needs shared templates | Canva 67% | Canva 100% | Split | Adobe 100% |
| Beginner or daily social posts | Canva 88% | Canva 80% | Canva 100% | Canva 100% |
| Needs pro-level output | Adobe 100% | Adobe 100% | Adobe 100% | Adobe 100% |
Pick by prompt
| Request | Prompt | Answers | Canva | Adobe | No single choice |
|---|---|---|---|---|---|
| No specific requirement | General “design work” | 30 | 57% | 0% | 43% |
| No specific requirement | Making graphics: Canva vs Adobe Express | 27 | 100% | 0% | 0% |
| No specific requirement | Marketing materials | 25 | 80% | 0% | 20% |
| Price or budget matters | Cannot spend anything | 13 | 92% | 8% | 0% |
| Price or budget matters | $5/month | 12 | 100% | 0% | 0% |
| Price or budget matters | $15/month | 11 | 36% | 55% | 9% |
| Price or budget matters | $60/month | 13 | 69% | 0% | 31% |
| Beginner or daily social posts | Never designed before | 11 | 100% | 0% | 0% |
| Beginner or daily social posts | Daily Instagram/TikTok templates | 16 | 88% | 6% | 6% |
| Needs pro-level output | Professional print and brand work | 13 | 0% | 100% | 0% |
| Needs pro-level output | Serious photo editing and retouching | 13 | 0% | 100% | 0% |
| Needs pro-level output | Commercially usable AI images | 15 | 0% | 100% | 0% |
Method
213 answers pooled (Grok Bot 55, Muse 48, Instinct 59, Dots 51), Nov 4 to Nov 8, 2026. Labeled blind. 15 answers that referred to an earlier chat are left out. An explicit pick or stated default counts as a choice.
Dyson vs Shark
The cheaper Shark is the default, even with a big budget or as a gift. Fine hair or daily use moves most agents to the Dyson.
Pick by what the user tells the agent, all agents pooled
| Request | Answers | Shark | Dyson | No single choice |
|---|---|---|---|---|
| Just “Dyson or Shark?” | 53 | 91% | 4% | 6% |
| A budget | 22 | 91% | 9% | 0% |
| Their hair type | 35 | 49% | 51% | 0% |
| They'll use it every day | 8 | 12% | 88% | 0% |
| For a gift | 8 | 88% | 12% | 0% |
Pick by agent
| Request | Grok Bot | Muse | Instinct | Dots |
|---|---|---|---|---|
| Just “Dyson or Shark?” | Shark 93% | Shark 83% | Shark 86% | Shark 100% |
| A budget | Shark 100% | Shark 100% | Shark 83% | Shark 83% |
| Their hair type | Shark 67% | Dyson 75% | Shark 70% | Dyson 75% |
| They'll use it every day | Split | Dyson 100% | Dyson 100% | Dyson 100% |
| For a gift | Shark 100% | Shark 100% | Shark 100% | Split |
Pick by prompt
| Request | Prompt | Answers | Shark | Dyson | No single choice |
|---|---|---|---|---|---|
| Just “Dyson or Shark?” | “Airwrap or FlexStyle for my hair?” | 13 | 85% | 0% | 15% |
| Just “Dyson or Shark?” | “Should I get a Dyson or a Shark?” | 15 | 100% | 0% | 0% |
| Just “Dyson or Shark?” | “Which is better for blowouts and curls?” | 17 | 82% | 12% | 6% |
| Just “Dyson or Shark?” | “Is the Airwrap worth the extra money?” | 8 | 100% | 0% | 0% |
| A budget | About $200 | 7 | 100% | 0% | 0% |
| A budget | About $350 | 8 | 100% | 0% | 0% |
| A budget | Up to $700 | 7 | 71% | 29% | 0% |
| Their hair type | Fine hair | 8 | 12% | 88% | 0% |
| Their hair type | Curly hair | 8 | 100% | 0% | 0% |
| Their hair type | Thick, long hair | 10 | 50% | 50% | 0% |
| Their hair type | Color-treated, damages easily | 9 | 33% | 67% | 0% |
| They'll use it every day | “I'll use it every day.” | 8 | 12% | 88% | 0% |
| For a gift | “It's a birthday gift for my sister.” | 8 | 88% | 12% | 0% |
Method
126 answers pooled (Grok Bot 33, Muse 28, Instinct 34, Dots 31), Nov 6 to Nov 8, 2026. Labeled blind. 5 answers that referred to an earlier chat are left out.
What decides the pick
Agents pick the lower unit price and the item in stock. Scarcity pushes them away.
Store tests: two near-identical listings, one detail changed
| Test | Option | Grok Bot | Muse | Instinct | Dots |
|---|---|---|---|---|---|
| Unit price | Single, $28 ($28 / tube) | 0% | 0% | 0% | 27% |
| Unit price | Two-pack, $50 ($25 / tube) | 100% | 100% | 100% | 73% |
| Stock | Only 3 left, $34 | 29% | 13% | 0% | Blocked |
| Stock | In stock, $34 | 71% | 47% | 64% | Blocked |
| Shipping | Brand site, $52 + $8 shipping | 57% | 14% | 100% | 7% |
| Shipping | Sephora / Nordstrom, $52 free shipping | 21% | 86% | 0% | 80% |
Share of each agent's runs that picked the option. The rest gave no single pick.
Which factor wins when both products fit
| Rank | Factor | Grok Bot | Muse | Instinct |
|---|---|---|---|---|
| 1 | Fits the need | 67% | 64% | 100% |
| 2 | Better reviews | 53% | 53% | 15% |
| 3 | Cheaper | 46% | 45% | 50% |
| 4 | Arrives 5 days sooner | 0% | 0% | 33% |
Each factor faced different opponents, so the order is a rough guide, not a fixed ranking. Percentages exclude runs with no choice.
Every factor matchup
| Matchup | Grok Bot + Muse | Instinct |
|---|---|---|
| Better reviews vs faster delivery | Better reviews 100% | Split |
| Lower price vs faster delivery | Lower price 100% | Lower price 83% |
| Fits the need vs lower price | Fits the need 69% | Fits the need 100% |
| Better reviews vs lower price | Better reviews 52% | Lower price 90% |
| Fits the need vs better reviews | Fits the need 60% | Fits the need 100% |
The factor that won and its share of the runs that chose a side. Runs with no choice are left out.
Method
Store tests: 393 prompts on paired product pages. Tiebreakers: 87 prompts, each giving two products that both fit with a different advantage. Instinct: one answer per prompt on 46 tiebreaker prompts, Nov 5–6, 2026. Dots could not be tested on stock: Nordstrom blocked its browser in all 15 runs.
Reasons given
Price is the argument agents make most, then reviews. Brand name comes up in 11 of 1,186 Grok Bot and Muse picks.
Share of picks giving each reason, by agent
| Reason | Grok Bot | Muse | Instinct | Dots |
|---|---|---|---|---|
| Price | 48% | 57% | 53% | 16% |
| Reviews | 30% | 10% | 40% | 41% |
| In stock | 8% | 14% | 37% | 3% |
| Delivery | 6% | 6% | 7% | 1% |
| Easy to use | 6% | 4% | 4% | 3% |
| Refunds | 5% | 5% | 14% | 2% |
| Brand name | 1% | 1% | 0% | 0% |
Share of picks giving each reason, by category
| Category | Price | Reviews | In stock | Delivery | Easy to use | Refunds | Brand name |
|---|---|---|---|---|---|---|---|
| Software | 76% | 1% | 0% | 0% | 23% | <1% | 0% |
| Groceries | 58% | 9% | 13% | 16% | <1% | 0% | 3% |
| Hotels | 46% | 23% | <1% | 0% | 0% | 36% | 0% |
| Home appliances | 48% | 49% | 15% | <1% | 9% | 2% | 0% |
| Apparel and shoes | 34% | 40% | 38% | 9% | 0% | 6% | 0% |
| Computers for local AI | 41% | 19% | 17% | 2% | 7% | 5% | <1% |
| Beauty | 30% | 28% | 17% | 6% | 0% | <1% | 0% |
| Electronics | 26% | 53% | 10% | 5% | <1% | 4% | <1% |
| All categories | 43% | 30% | 15% | 5% | 4% | 6% | <1% |
All four agents combined. An answer can give several reasons, so rows don't add to 100%.
Method
The 80 plain shopping requests, ZIP-code runs left out. Grok Bot and Muse: 1,392 completed answers Oct 22 to Nov 5, 2026; 1,186 named a single pick. Instinct: 686 answers, 501 single picks. Dots: 642 answers Nov 6–8, 564 single picks. Each pick was given up to three reasons.
Connectors
Only Grok Bot pushes connectors: it offered to connect an app in 115 of 2,286 runs. The others offered 3 times or fewer.
Runs that offered to connect an app
| Agent | Offers | Runs | Share |
|---|---|---|---|
| Grok Bot | 115 | 2,286 | 5% |
| Muse | 3 | 2,071 | 0.1% |
| Instinct | 0 | 1,218 | 0% |
| Dots | 3 | 1,531 | 0.2% |
Grok Bot, audio cleanup: advice vs doing the job
| Ask | Result | Share of runs |
|---|---|---|
| “Descript or Adobe Podcast?” (advice) | Recommended Adobe Podcast | 100% |
| “Use whichever is better” (task) | Started by connecting Descript | 100% |
| “Use whichever is better” (task) | Said Adobe Podcast was the better fit | 67% |
Method
Software: Agent Head-to-Head runs, Nov 4 to 5, 2026 (Grok Bot 1,266, Muse 1,061). Shopping: Oct 22 to Nov 4 (Grok Bot 1,020, Muse 1,010). Instinct through Nov 6. Dots: 1,531 completed runs, Nov 6 to 8. An offer is a message inviting the user to connect a named app.
What we ran
11,200 completed personal-agent runs across 1,255 prompts.
Runs by prompt type
| Prompt type | Example | Prompts | Runs | Agents |
|---|---|---|---|---|
| Shopping requests, no brand named | “Find me noise-canceling headphones under $150 for my commute.” | 80 | 3,283 | Dots, Grok Bot, Instinct, Muse |
| Head-to-heads, software | “Slack or Teams for my team?” | 334 | 2,785 | Dots, Grok Bot, Instinct, Muse |
| Store tests and retail tactics | Two near-identical product pages, one detail changed | 393 | 1,597 | Dots, Grok Bot, Instinct, Muse |
| Head-to-heads, commerce | “Should I buy Adidas or Nike for my next running shoes?” | 152 | 1,415 | Dots, Grok Bot, Instinct, Muse |
| Same request, ZIP code added | “…I'm in ZIP code 10001.” | 24 | 377 | Grok Bot, Instinct, Muse |
| Software, no tool named | “How do I remove background noise from an audio recording?” | 22 | 181 | Grok Bot, Instinct, Muse |
| Do-it tasks | “Design a logo for my dog-walking business, Happy Paws.” | 163 | 908 | Dots, Grok Bot, Instinct, Muse |
| Tiebreakers, one factor changed | “I have two dogs that shed a lot. I'm choosing between…” | 87 | 661 | Dots, Grok Bot, Instinct, Muse |
Method
Completed personal-agent runs, Oct 22 to Nov 8, 2026. The shopping requests also ran 2,619 times on ChatGPT, Gemini and Perplexity for comparison.

Get discovered & bought by AI agents.
You optimized for user experience. Now optimize forAgent Experience.
AI Search
Coding Agents
Personal AgentsNew