Representative AI Prompt Library: Aleyda Solis’s Method to Measure Your Real E-Commerce Visibility
Summarize this article with AI
When Your AI Visibility Drops from 62% to 11% by Changing Prompts
A Marketing Director calls me on a Thursday. He’s holding a report from a major AI tracking tool. His visibility score: 62%. He’s confident. I ask him a question: « Do your customers type ‘best wireless headphones’ or ‘headphones for running in the rain, delivered before Saturday’? » He hesitates. So we run a test. We replace 20 generic prompts with 47 prompts representative of his real customer journey. Result: visibility drops to 11%. Not an error. A revelation.
Aleyda Solis, international SEO consultant, says it loud and clear:
« One of the biggest mistakes I see with AI search tracking is using a non-representative prompt library. »I see it every week. Default tools use generic prompts: « top 10 », « best », « review ». These prompts don’t capture long conversational questions, delivery constraints, local comparisons, purchase doubts. Your dashboard shows generous presence. It matches no purchasing decision. The key? Build a custom prompt library. Not to track every possible query. To sample the journeys that matter. So, ready to see your real visibility?
Why Generic Prompts Distort Your Measurement
An AI query is not an SEO keyword. It’s long, chatty, contextual. A buyer doesn’t ask « wedding dress ». They write: « bohemian wedding dress for beach, $800 budget, delivery to Corsica before June ». If your prompt library ignores these constraints, you’ll never know your local competitor systematically appears on fast delivery.
Aleyda Solis warns: the goal is representativeness, not exhaustiveness.
« Representative doesn’t mean exhaustive. »Most e-commerce brands track 15 to 30 generic prompts. They miss the questions that trigger a purchase. I’ve verified it with 12 clients: moving from a generic list to a representative sample, the appearance rate on the 20% most transactional prompts dropped by half. In real life, these prompts are the ones your customers type at 10pm, in bed, before hitting checkout. To capture them, you must map the customer journey: discovery, comparison, purchase, support. And inject real constraints: budget, urgency, location, product attributes. This is the first step of Aleyda’s method: start from business questions, not prompts.
The Prompt Matrix: Aleyda Solis’s Tool for a Representative Sample
How do you build this sample without spending two months on it? Aleyda proposes a matrix. On one side, customer journey stages: Inspiration, Évaluation, Purchase, Post-Purchase. On the other, intention types: product information, comparison, transaction, support. You cross them. You add segmentation layers: market, product line, persona. You get a grid.
I used this matrix for an eco-responsible clothing brand. 3 markets (France, Belgium, Switzerland), 2 lines (organic cotton, linen). Here’s what one slice of their matrix looked like:
| Stage | Type | Real Prompt |
|---|---|---|
| Évaluation | Comparison | « Linen t-shirt vs organic cotton, which holds up better after 10 washes at 30°C? » |
| Évaluation | Information | « Does [brand] use heavy-metal-free dyes, GOTS certified? » |
| Purchase | Transaction | « Order a batch of 3 linen t-shirts, pickup point delivery in Brussels before December 20 » |
| Post-Purchase | Support | « My t-shirt shrank on first wash, what do I do? » |
Every prompt comes from real customer language: reviews, support tickets, Search Console queries, FAQs. Nothing invented. Aleyda emphasizes step 7: « Use real audience language from multiple sources ». It’s conversational language, not a keyword list. Once the matrix is set, you build the library by prompt groups, not one by one. A single need can be expressed 4 ways. You create variations: shorter wording, budget amount added, competitor mentioned. You get a solid sample without formulation bias. Total for this client: 47 prompts, not 200. Enough to see trends.
From 47 Prompts to +42% Organic Clicks: A Fashion E-Commerce Case
47 prompts. 6 months of tracking. Result: the brand’s AI visibility increased 42% on matrix prompts. Before optimization, it appeared in 14% of responses generated by ChatGPT and Perplexity. Today, 20%. +42%. No ads, no media budget. Just content tailored to the right questions. Organic traffic jumped 37%. Pages optimized for these AI prompts also capture long-tail queries on Google. That’s the AI-organic synergy.
Here’s how. I followed Aleyda’s protocol in four phases: (1) Define the matrix with the client, (2) Generate prompts from real sources, (3) Measure every two weeks, platform by platform, (4) Tag every prompt with a label: journey stage, intention type, competitor cited, market, language. Example prompt sheet:
- Prompt: « What linen t-shirt brand offers a straight fit, not too long, for a 5’9″ man? »
- Metadata: Évaluation, Product Information, Competitor X, France, Linen
- Platform: ChatGPT
- Result before: not cited
- Result after: cited in 3rd position with link
This tagging matters. Without it, you won’t know why you’re progressing. Aleyda repeats it at step 17: « Tag every prompt with metadata ». This spots when a competitor wins on eco-delivery questions, or your product sheet lacks specific info. Then you fix it. With this client, 30% of prompts pointed to a competitor because of missing product specs. We enriched sheets with the missing data. In 8 weeks, appearance rate on these prompts went from 0 to 35%. The content already existed on site, it was just poorly structured. We rebuilt it using Guillaume Attias’s DOSE framework (BMO Academy): semantic clusters around each product line, with comparison pages, FAQs, and technical content. Result: AI reads our pages like a knowledge base. It cites, it recommends. Organic flow follows.
Defensive Measurement: The Protocol That Turns Prompts Into Decisions
Measuring without protocol is like thinking a thermometer tells you your body temperature without calibrating it first. Aleyda devotes part 3 of her guide to this. A defensive protocol rests on three pillars. 1. Test by platform, without aggregating scores. ChatGPT isn’t Perplexity. Sources differ. Merging creates noise. 2. Define a frequency and detection threshold. For our fashion brand, we measured the 47 prompts every two weeks. A change under 5 points wasn’t significant. 3. Tag every prompt with the metadata I mentioned. Without it, impossible to see one product line progressing and another stalling. I add a fourth pillar: validate the sample before scaling. Aleyda calls it step 18: « Validate the prompt library before scaling ». You run the 47 prompts once, analyze if results make sense, verify they cover critical journeys, remove off-topic ones. Then you freeze the sample for three months. Many brands change prompts constantly. They track nothing, they flutter around. A stable sample is a stable mirror. I list the metadata we use without exception:
- Prompt ID
- Journey stage
- Intention (Information, Comparison, Transaction, Support)
- Competitor mentioned
- Market / language
- Product line / product
- Constraint (budget, urgency, location)
- Language source (FAQ, reviews, Search Console)
With this tagging, every measurement becomes a decision. You don’t ask « Did our visibility change? » anymore, you know « Which competitor is beating us on which delivery questions in Belgium ». That’s precise, actionable. And it takes no more time than fuzzy reporting.
From Prompt to Action: What You’ll Fix on Your Pages
The prompt library isn’t an end in itself. It’s a radar. Once you’ve spotted the gaps, you take action. For our client, signals dictated three priorities. 1. Enriching product sheets: on 30% of prompts, AI missed our products because technical attributes weren’t structured in readable text. We added descriptive paragraphs with key data: composition, wash, label, fit. 2. Building comparison pages: prompts like « X vs Y » needed neutral, detailed content. We built 22 semantic clusters around comparisons. Result: AI started citing our brand as neutral reference. 3. Conversational FAQ: post-purchase prompts (returns, washing, sizing) showed a gap. We published FAQ in natural language, matched to real customer messages. In 6 weeks, AI citations on these topics tripled.
Each action was directly tied to a library prompt. This is the missing link in most AI stratégies. You measure, discover, then fix. Aleyda closes her guide with step 20: « Connect prompts to the optimization workflow ». Without this link, you pile up useless data. With it, you turn every visibility percentage into an editorial task. And the virtuous cycle kicks in. A representative prompt library, refreshed quarterly, feeds your SEO roadmap. It aligns your content and SEO teams. No more debates on « should we write about this keyword? ». The question becomes: « What prompts are our customers asking this month? ».
So, does your prompt library capture the customer hunting fast delivery at 11pm on a Tuesday? Or does it just measure « top 10s » that don’t sell?
Your AI Visibility Audit in 47 Minutes
I build a representative prompt library for your e-commerce with you, live. We spot ignored customer journeys, tag prompts, run an initial measurement. Zero PowerPoint. Pure action.
Book a strategic call — 45 minFrequently Asked Questions
Why aren’t the default prompts from AI tracking tools enough?
Generic tools give one-size-fits-all prompts. They ignore your buyers’ real constraints: budget, location, specific comparison. A library built on your customer journeys offers concrete visibility.
How many prompts do you need for a representative library?
You don’t need hundreds. Aleyda Solis recommends a minimum viable sample. Depending on the case, 30 to 70 well-chosen prompts are enough to detect trends. Representativeness matters, not volume.
How do you find real customer language to write prompts?
I analyze support tickets, reviews, social media questions, Search Console data, and chatbots. That’s a goldmine of natural phrasing. Aleyda puts it well: « Use real audience language from multiple sources ».
Should you measure across all AI platforms (ChatGPT, Perplexity, Google SGE)?
Yes, but separately. Each platform has its own sources and algorithms. Merging scores muddies the picture. Measure each platform, then cross-reference trends.
How often should you update the prompt library?
Every quarter. Add or remove prompts based on new customer questions, product launches, or market shifts. A frozen library quickly becomes outdated.
