Case study · 29 June 2026

Spaartje passes the free snapshot. The full audit tells a different story.

Spaartje is an AI grocery-price-comparison app for the Netherlands: 13 supermarkets, ~140K products, free. Their site is one of the most AI-ready in the country. Every crawler allowed, a model-ready llms.txt, fourteen schema types on the homepage. We ran the free Far & Wide snapshot againstspaartje.comin 30 seconds. Four passes, zero warnings, zero fails.

Then we ran the full audit. It returned numbers the snapshot could never have predicted.

Layer 1 · The free snapshot

Everything an AI engine could ask for, technically.

Snapshot conclusion: Spaartje is the textbook example of a site doing the technical basics of AI visibility right. Most clients arrive at this layer with at least one warn or fail. This is the bar.

Layer 2 · Presence (what the full audit found)

Passing every check doesn't mean AI recommends you.

We ran 100 verified Dutch and English customer questions through ChatGPT, Gemini, Perplexity, Claude, and Google AI Overviews. All five AI engines. Location and language set to match what a real user sees.

EngineMentioned in AI answers (first run)After language fixTone of answers
Gemini34%67%Neutral
Claude30%—Positive
Perplexity29%39%Positive
ChatGPT (API)4%0%Negative

The headline. Where Spaartje appears, it appears first (average position #1 on every engine) and the tone skews positive. Perplexity recommends it in 68% of mentions. On ChatGPT's bare API it is essentially absent. That does not mean it is absent in the ChatGPT app: a paired test ranked Spaartje #1 above three competitors on a Dutch phone. It means the measurement tier matters, and the snapshot alone would never have shown that.

The root cause, surfaced by Layer 6 (Consensus): zero independent domains name Spaartje. Their own site is the only source AI has to draw on. Engines synthesize across sources. With nothing to triangulate, they hedge or stay silent.

The fix plan

Five moves, ordered by impact and effort.

The full audit produces a prioritised fix list. Here is what Far & Wide generated for Spaartje. Each fix names the layer it addresses and the exhibit that backs it up.

  1. #1Layer 6 · Consensus
    Impact H·Effort M

    Get cited by the category-authority domains AI already trusts

    Today: 0 independent domains name Spaartje. Engines have nothing to triangulate. The supply-chain analysis surfaces the domains AI cites for this category (supermarktscanner.nl, gierigegerda.nl, consumentenbond.nl, bright.nl). Earned mentions on 2–3 of these lift your mentions across every engine, not one.

  2. #2Layer 5 · Accuracy
    Impact H·Effort S

    Lock the supported-store list as one machine-readable source of truth

    Today: store count appears as 12 / 13 / 14 across site, App Store, Play Store, press. Engines repeat all three. Picnic is in or out depending on which surface they read. One canonical /supported-stores.json mirrored into the homepage + llms.txt collapses the entire ‘needs client confirmation’ cluster of 36 flagged claims.

  3. #3Layer 4 · Citation
    Impact H·Effort L

    Earn citation alongside the aggregators, not against them

    Today: spaartje.com is the #3 most-cited domain in this category, but the top two are aggregator indexes and competitor sites (supermarktscanner.nl 55× · groceriestracker.com 23× · pingprice.be 21×). Pitching to be listed on each of those flips earned citations from 0 toward parity.

  4. #4Layer 3 · Preference
    Impact M·Effort S

    Reinforce the differentiator in the homepage hero: full basket, not generic deals

    Today: where Spaartje appears it appears first (avg position #1), but several answers describe it as a generic ‘deals app’ rather than the full-basket comparison it is. The homepage opening line should carry the differentiator literally: ‘compares your full grocery basket across 13 Dutch supermarkets and shows the cheapest shopping plan.’ Engines repeat what they read.

  5. #5Layer 2 · Presence (ChatGPT specifically)
    Impact H·Effort S

    Measure ChatGPT as customers use it: the bare API is a floor, not the truth

    Today: asking the ChatGPT API directly reports 0% mentions. ChatGPT as customers use it (with NL location and the app layer the API does not expose) ranked Spaartje #1 above three named competitors in a paired test. The fix is not on Spaartje's side. It is how the measuring is done. The full audit re-asks the failing questions in a real logged-in ChatGPT session, with screenshots as evidence, so the real number gets captured.

Want to see this for your own brand?

The snapshot is free and takes 30 seconds. The full audit has the same shape as Spaartje's above: your engines, your questions, your fix plan.

Get my free snapshot →See pricing