For retailers and dealers / Evidence
Every figure we use, and what it actually says.
We sell decision integrity, so it would be strange to make a claim you could not check. Figures are stated as their source states them. Where a comparison is self-selected or the sample is a subset, that is noted rather than hidden.
| Figure | Category | What it actually says | Source |
|---|---|---|---|
| 13:13 → 13:59 | Vehicles | Total time spent shopping for a vehicle, 2020 against 2025. Effectively flat straight through the AI wave. Roughly seven hours of it happens online. | Cox Automotive Car Buyer Journey Study, 2025 |
| 46% / 24% | Cross-category | Among AI users, the share who start product research with AI against the share who start with search. Not a figure about all consumers. | L.E.K. Consulting, Apr 2026, n=2,650 |
| 5% | Cross-category | Would act on an AI recommendation above $500 without verifying it elsewhere. | Product.ai, Apr 2026, n=1,463 |
| 60% / 63% | Cross-category | Abandoned a purchase because there were too many options, and say purchase decisions take more effort than they used to. | Google / Ipsos, Mar 2024, n=1,000 |
| 48% / 23% | Cross-category | AI-referred visits last 48% longer and convert 23% worse. More time, worse outcome. | Adobe Analytics, Mar 2026 |
| 81% / 65% | Vehicles | Satisfaction among buyers who used AI in their search against those who did not. A self-selected comparison, not the same buyer measured twice. | Cox Automotive, 2025 |
| 74% / 48% | Vehicles, NZ | NZ new-car buyers who rule out an EV on battery lifespan and resale value, and who contact only one dealership. A national average you cannot cut by model, region or dealer. | Trade Me Motors, Aug 2026, n=1,949 |
| 0.2 pts / 7.4 pts | Appliances | Two big-box appliance retailers draw within 0.2 points of each other and sit 7.4 points apart on unit share. The gap is built almost entirely after arrival. US category data, presented as category physics rather than local numbers. | OpenBrand / TraQline, Q1 2026 |
| 75.4% | Appliances | Share of major appliance purchases completing in store, up 1.8 points. | OpenBrand / TraQline, Q1 2026 |
Questions
The things people ask before a pilot.
Is this the same as answer engine optimisation?
No, and the two work at different ends. AEO competes to be the source an AI summarises. The content engine produces exactly the sort of material that gets summarised, so you benefit there as a side effect. But the buyer who has read the summary still has a decision to make, and that is the part AEO cannot help with. The interface is for what happens after the answer.
Why would a buyer use a decision tool on a retailer's own site?
Because the ranking moves when they move their priorities, and they can see exactly why. The interface reasons with them rather than handing them a conclusion. On your site the range is yours and the buyer knows that, which is why the independent surface on pyfoi.com has to stay genuinely independent.
Who writes the content, and how do you produce that much without it being wrong?
Pyfoi produces it through a pipeline where competing models attack each draft before a human reviews it, and nothing publishes until it clears three gates: fact check, misleading claims and legal, and brand. Claims trace back to a source the buyer can open. On your own site, your people approve every piece of brand response before it goes live.
We already have a content team. What changes?
Cadence and coverage. Most category content answers the questions a brand is comfortable answering, at the rate a team can write them. This answers the full mapped set, including the drawbacks and the comparisons that do not flatter anyone, and it re-cuts within weeks when a new objection starts forming rather than after the next annual survey. Your team stays in the loop as the approver rather than the bottleneck.
Who owns the decision data?
You do. Consent sits with you. We do not market to your buyers and we do not share their data with anyone else. What improves across the platform is pattern-level only.
Does Pyfoi sell ranking positions?
No. On pyfoi.com the ranking covers every brand in the category, including products a partner does not stock, and no partner can influence it. Traffic from pyfoi.com to a partner site is free. Partners pay for the decision data and for the interface embedded on their own site.
What does a pilot actually involve?
One category, one market, a mutual NDA, and a baseline taken from your existing analytics rather than ours. The interface goes live on a defined page set and is measured against that baseline on decision completion, return visits and lead quality.
Does it work for appliances and electronics as well as vehicles?
Yes, though the evidence differs and we keep the two things separate. Vehicles is the only category with a clean multi-year series on research hours. For appliances the equivalent instrument does not exist publicly, so the argument runs on draw against close: two retailers pulling identical traffic and finishing seven points apart on share. Same physics, different measurement. Each category has its own page with its own numbers rather than one page with the figures swapped out.
What about real estate, financial services or health products?
The engine transfers. The constraints do not. In financial services the constrained speaker is Pyfoi rather than you, which changes what the independent surface can say and means no ranking there at all. In health supplements there are no products on pyfoi.com. Worth a conversation rather than a page.