LSLLM SHELF
View reportPLBook a call

Win
AI recommendations

Your potential customers bring their problems to AI. Your product may be the right solution,
but does AI see it that way?

AI does more than recommend.

It can also advise against products.

An in-depth audit of your AI visibility

Rigorous analysis, not just a handful of numbers.

Differences between models

Brand recommendation rate
13.8%
Google AI
Overview
11.1%
Gemini
4.4%
ChatGPT
3.1×difference between the highest and lowest result

The product wins in selected buying situations

Recommendation rate by user need
Feeding format
40.6%
Formula and claims
9.3%
Health
5.9%
Fussy eater
2.3%
General choice
0%
4 / 7buying situations end with a brand recommendation

Question type matters

Brand activation by intent
Format switch
72.2%
Format selection
51.9%
Claim evaluation
27.8%
Myth correction
12.5%
Discovery
0%
72.2%format-switch questions activate the brand

Head-to-head: who really competes

Comparison only within the same responses
COMPETITORCO-OCCUR.BRANDCOMPETITORWIN RATE
2882028.6%
2221195.5%
127558.3%
97277.8%
Direct comparisons reveal the biggest threat.
Across 28 shared responses, the brand won 8 times and lost 20.

Sources: where the leader’s advantage is built

Domains associated with category wins
314losses
27present, not #1
287absent
286losses with sources
SOURCEWINSSHAREMODELS
2140.4%2
1630.8%3
1528.8%3
91.1%losing responses included sources

Unbranded vs follow-up

How intent changes after the first response
Skyr — start
68%
Skyr — follow-up
76%
Bar — initial
36%
Bar — follow-up
57%
Retailer — follow-up
39%
The next question deepens purchase intent.
The follow-up reveals expectations around product, brand, price and availability.

What we measure
and interpret

We use multiple metrics to identify the challenge precisely and determine how to address it.

Brand visibilityHow often the brand appears in model responses.
Recommendation rateShare of responses containing a recommendation.
Brand positionHow high the brand appears on the shortlist.
Strength of evidenceThe depth and quality of arguments supporting recommendations.
StabilityWhether the result holds across repeated runs.
Response coverageHow broadly the brand covers different questions and contexts.
Presence consistencyWhether the brand appears consistently across different situations.
User intentsWhich intents and needs activate the brand.
Activation typeWhether the brand enters through discovery, ranking or follow-up.
Use casesScenarios in which the brand wins or disappears.
Formula and claimsWhich claims and product features support the recommendation.
Impact of web access and sourcesHow web access or a specific domain changes the shortlist, reasoning and recommendation.
Recommendation conditions and caveatsConditions, reservations and limitations in model responses.

AI is not search

The journey does not end after one query.
It is a conversation, and every follow-up can change the shortlist and recommendation.

ChatGPT 4o⌄Follow-up scenarios
Recommendation recovery
I am looking for good food for a dog with a sensitive stomach.
Look for an easy-to-digest food with a single protein source and a simple formula.
I would consider Purina Pro Plan Sensitive Digestion — as the first trial, with Hill’s Sensitive Stomach & Skin — and Royal Canin Digestive Care —.
Decision narrowing
I have combination, acne-prone skin. Which simple active product should I consider?
Products with salicylic acid, azelaic acid or niacinamide are often useful. Options include Paula’s Choice 2% BHA Liquid, Acne-Derm 20% or The Ordinary Niacinamide 10% + Zinc 1%.
For a gentle start, I would choose The Ordinary Niacinamide 10% + Zinc 1% — because it is simple to use; for active acne, it is worth consulting a dermatologist.
Basket expansion
I need a lightweight laptop for work and travel.
A good choice would be ASUS Zenbook 14 OLED UX3405 — it is lightweight, has a good display and can handle many browser tabs.
Yes: Logitech MX Anywhere 3S — and a compact 65 W USB-C charger. I would not buy an extra stand without a specific need.
Anti-recommendation
I am looking for white sneakers for the office and everyday wear.
Look for a leather upper, subtle branding and an easy-to-clean sole.
I would avoid Nike Air Force 1 Shadow — if you want a discreet silhouette, and Adidas Ozweego White — if the shoe also needs to work with more formal outfits.
＋　Ask anything♩　▮▮
Simulated conversations. AI responses may contain errors.

The persona changes the recommendation

The same broad buying task produces different answers when the user’s age, goal, budget, lifestyle or constraints change.

ChatGPT 4o⌄Different personas
Busy professional
I have many meetings and need a healthy mid-morning snack that I can buy on the way and eat without preparation.
The most practical option is Piątnica Skyr pitny — it is portioned, widely available and needs no cutlery. Add fruit if you need a more filling combination.
Post-workout user
I am 22, returning from strength training and want to replenish protein and energy quickly.
A better choice would be Olimp Whey Protein Complex 100% — with a banana or another carbohydrate source, rather than a small yoghurt alone.
Parent of a teenager
I need something for a teenager’s backpack: convenient, filling and not too sweet.
I would consider Piątnica Skyr naturalny — with a piece of fruit. This combination is easier to match to appetite than a ready-made sweet dessert.
Senior
I am 72, have a smaller appetite and need a mild, small meal with protein.
I would choose Piątnica Serek Wiejski naturalny — in a small serving. If you have dietary restrictions or chronic conditions, confirm the choice with a doctor.
Budget-conscious student
I need a cheap, filling mid-morning snack. I have access to a fridge but do not want to cook.
The best value for satiety comes from Pilos Skyr naturalny — with oats and a banana; a ready-made bar will usually cost more per serving.
＋　Ask anything♩　▮▮
Simulated conversations. The recommendation depends on the information given to the model.

How
LLM Shelf works

We tailor the analysis to your needs and focus on what matters most to you.

Priority map

We assess several areas and identify the ones worth exploring in depth.

Strategic report

We analyse use cases, models, follow-ups, KPIs, sources and recommendations.

Insights, recommendations
and implementation

We turn data into an action plan
and support the team with implementation.

Impact measurement

We verify whether implementation actually changes AI recommendations.

Let’s see whether AI recommends your products or your competitors’.

Order an audit

Frequently asked questions

How an AI visibility audit works for brands and products.

What does brand visibility in AI mean?

It is more than a single mention. It includes brand recognition, spontaneous recall by the model, entry into the shortlist, position versus competitors and the final recommendation.

Do you measure only brand visibility, or can you also measure product visibility?

From a user's perspective, the brand is often secondary to the need, so we also measure product visibility. We agree the research granularity with the client: it can cover a brand, a product line, a specific product or even an individual SKU.

What exactly does LLM Shelf measure?

We measure brand or product visibility and recommendation rate, shortlist position, strength of supporting arguments, result stability, differences between models, performance across buying situations, what happens after users ask follow-up questions, and the impact of web access and specific sources.

How is an AI visibility audit conducted?

We build a question set around real customer needs, test multiple models and repeat the tests. We analyse first answers, follow-ups, personas, sources, direct comparisons with competitors, and variants with and without web access.

Why is one question to ChatGPT not enough?

Models are probabilistic: the same question can produce different answers. A single screenshot is an observation, not a reliable measurement. Repetitions, question variants and a clear aggregation method are required.

Which AI models do you analyse?

The scope depends on the market and research objective. An audit can include the most widely used environments: ChatGPT, Gemini, Google AI Overviews, Perplexity, Claude, Microsoft Copilot, Grok, DeepSeek and Meta AI.

How is AI visibility different from SEO?

SEO mainly asks whether a page can be found. LLM Shelf checks whether the information AI finds makes it consider a brand or product, place it on a shortlist and choose it over competitors. SEO is a foundation, but it is not the whole recommendation process.

Can AI know a brand but still not recommend it?

Yes. Brand knowledge can open the door, but recommendation also depends on fit with a specific need, current information, evidence quality, credibility and advantage over alternatives.

Can you identify the exact page that caused a recommendation?

Not always with certainty, because closed models do not reveal their weights or full decision process. We can compare answers with and without web search, with and without a specified source, and test how a particular domain changes the shortlist, reasoning and final recommendation.

What does a company receive after an audit?

A diagnosis of the brand and product position, a competitor benchmark and a prioritised action plan. Recommendations may cover technical accessibility, information architecture, buying-situation content, claims, external evidence and conflicting or outdated information.

Can you guarantee that AI will start recommending a brand?

No independent probabilistic system can be guaranteed to produce a specific answer. We can reliably measure the baseline, identify likely barriers, implement changes and test whether the result improves.