New Aleyda Solis workbook: your AI visibility in e-commerce doesn’t depend on your prompts. It depends on their representativeness.
Summarize this article with AI
Your AI prompt library may be the weak link in your measurement
I audit 15 e-commerce sites per week. The same flaw shows up in 12 audits.
An AI Visibility dashboard climbing. Screenshots of ChatGPT where the brand appears in responses. The boss thinks we’re well positioned on AI.
But the dashboard rests on 42 prompts. 80% are brand queries. The remaining 8 are generic keywords typed with no purchase constraint. Result: the metric gives a false picture. AI organic traffic isn’t taking off. The brand is invisible on real purchase journeys.
Aleyda Solis, international SEO consultant and creator of the SEOFOMO newsletter followed by more than 45,000 professionals, makes the diagnosis right from the first line of her new workbook published June 17, 2026:
« One of the biggest mistakes in AI Search tracking is using a non-representative prompt library. »
It’s a brutal field observation. The problem is there.
The workbook doesn’t reinvent AI tracking. It rewires it. It forces you to define what a sample of prompts should represent, before you even write the first prompt. In e-commerce, that changes everything.
Avant d’appliquer la méthode du workbook, le client suivait 57 prompts dont 48 contenaient sa marque. Résultat : un faux sentiment de visibilité IA. Voici la répartition réelle.
Typical AI tracking library: 84% brand queries
Répartition des 57 prompts d’un site e-commerce avant correction
He thought he was measuring AI visibility. He was only measuring his brands
A client in bathroom fixtures, 2,400 SKUs. He invested in an AI tracking solution as early as January 2025. His dashboard showed a B visibility score and an upward trend. But AI-assisted sales weren’t taking off.
I connect for a direct audit. I ask him to extract the list of tracked prompts.
57 prompts. 48 contained his brand name or one of his flagship products. 6 were generic questions like « best thermostatic mixer tap ». 3 top lists copy-pasted from Google Search Console.
No prompt mentioned budget. None mentioned location. None mentioned delivery time. None described a real problem (« too little water pressure in the shower »).
The dashboard said B, but the site was absent from 90% of conversational purchase journeys.
We redid everything with the workbook method. We expanded to 186 prompts. We cut the brand prompt count in half. We injected purchase constraints, local variants, expressions from support tickets.
Result: AI visibility that reflects reality. And exploitable gaps where the site wasn’t citing the right content.
Le workbook d’Aleyda Solis structure la construction d’une bibliothèque de prompts en 21 étapes réparties en 4 blocs. Voici les phases clés pour garantir la représentativité.
Les 21 étapes du workbook en 6 grandes phases
Du cadrage des questions métier au cycle de mise à jour
What the workbook changes: moving from sampling to reflecting the purchase journey
Aleyda Solis repeats it: representative doesn’t mean exhaustive. We don’t track every query. We build a structured sample that captures AI-assisted journeys that matter to the business.
The workbook details the method in 21 steps, divided into 4 blocks.
1. Define what the library should cover – the first 5 steps force you to ask business questions before writing a single prompt. Which markets? Which customer segments? What site type (brand, marketplace, distributor)? At what stage of the purchase journey?
A cross-tab, the « prompt matrix », forces you to visualize the intersections. No moving forward if the matrix is empty.
2. Build the prompt set – 6 steps that integrate real constraints: budget, location, delivery time, material, compatibility. Prompts aren’t simple questions. They’re spoken sentences, often long, with clear purchase intent. Aleyda insists: « Use real audience language from multiple sources ». We don’t guess, we collect from live chat, customer reviews, support, forums.
3. Measure without cheating – 6 steps to separate tests by platform (ChatGPT, Google AI Overviews, Perplexity), set the sampling protocol, tag each prompt with structured metadata.
4. Turn results into concrete actions – 3 steps to link each prompt to a page or content type, define update frequency and classify prompts as « core », « experimental » or « monitoring ».
21 steps. No scatter. A framework that prevents a flattering but false dashboard.
Adapting the library to an e-commerce site: purchase constraints, local variants, competitors
In e-commerce, representativeness easily falls into the catalog trap. You think adding « buy », « price », « reviews » is enough. The workbook gets you out of that illusion.
Take step 6: add real purchase constraints. For a lighting site, « design pendant » isn’t enough. You need « brushed brass pendant for 2.5m ceiling with E27 socket, $120 budget, ships in 4 days ». A prompt like what a user actually types in ChatGPT or Perplexity when they’re furnishing their living room.
Step 11: localize by market, beyond language. A Quebec prompt doesn’t match a French prompt. Expressions, postal codes, competing brands, electrical standards differ. On a site selling in French-speaking Belgium, I had to create 31 additional prompts just to reflect the local retailers mentioned by AIs.
Step 15: balance brand, non-brand and competitor prompts. The workbook recommends calibration: if your library is 60% brand queries, you’re measuring awareness, not visibility. My distribution client went from 85% brand prompts to 30%. The visibility score dropped mechanically. But it became true. And actionable.
Step 19: connect each prompt to an optimization lever. A prompt that surfaces a competitor instead of your brand should point to a product sheet, a guide page or missing structured data. The workbook formalizes this loop: prompt → gap → action → measurement.
From prompt to optimization: a loop you can action tomorrow
The real value of the workbook is it doesn’t stop at tracking. It arms you to turn an indicator into a concrete SEO project.
Aleyda offers a « quick check » at the end of the workbook. I use it systematically before validating a library. Five questions: do the prompts cover at least one complete journey? Do they contain real constraints? Are they split by platform? Are they tagged? Are they reviewed every 60 days?
At the distribution client, the library is now cycled on 90 days. We remove prompts with no conversational volume, we add ones from support verbatims. The dashboard doesn’t lie anymore. It shows real gaps.
Example: of the 186 prompts, 17 mentioned a specific faucet type missing from our content. We produced 4 technical guides in 3 weeks. Citations went from 0 to 13 in a month. Not a prediction. An observation.
Representativeness isn’t optional. It’s the foundation of measurement.
Aleyda Solis’s workbook sets a rigor standard for anyone who wants to steer AI visibility without fooling themselves.
In e-commerce, purchase journeys are long, multi-device and conversational. A prompt library built on brand and head terms delivers nothing. It creates the illusion of presence, it delays optimization.
The 21 steps are the minimum to avoid wasting 6 months tracking on false signals.
I often hear you say: « We’re invisible on AI ». The question is: with which prompt library are you measuring that?
Audit your AI prompt library in 90 minutes
I don’t hand you a standard 12-month report. I show you live how 21 steps transform 40 decorative prompts into a set of 180 prompts that truly reflect your purchase journeys. And I give you the key matrix ready to use.
Book a strategic call — 45 minFrequently Asked Questions
What’s the difference between a simple keyword list and a representative prompt library?
A keyword list assembles short queries. A library reproduces conversational journeys, with purchase constraints, local variants and natural formulations from customers. It shows the real purchase journey, not the catalog.
How many prompts minimum does an e-commerce site need?
No absolute number in the workbook. Start with a sample that covers at least one complete journey per product line, market and customer segment. Example: for a single-country site with 5 product ranges, 80 to 120 prompts are enough. Multiply by market and segment.
How do you balance brand and non-brand prompts?
Aleyda Solis recommends not exceeding 30% brand prompts. This avoids masking real gaps on discovery or comparison queries. If you go higher, your dashboard only shows your awareness, not your AI presence.
How often do you update a prompt library?
I align frequency with seasonality. 90 days for e-commerce with renewed collections, 60 days for competitive markets. My monitoring prompts (new AI use cases), I re-evaluate every 30 days.
Is Aleyda Solis’s workbook free?
Yes, you can download the workbook free from her site (aleydasolis.com), updated June 17, 2026. It compléments an article that details all 21 steps and explains the representativeness logic.
