Y1 Digital · Academy · Aug 2026

AI Visibility

How brands and their products get found when nobody searches anymore

40 model answers · pink: the ones that name your brand

Cold open. The grid is the pitch: forty sampled model answers, and only the pink ones name the brand. Visibility is a rate, not a ranking.

Users are already there.

0%

of US online shoppers used an AI assistant for product research in the past 90 daysCapital One Shopping Research, 2026

+0%

AI-referred traffic to US retail sites, Q1 2026 vs Q1 2025Adobe Analytics

+0%

better conversion from AI-referred visitors than non-AI trafficAdobe Analytics, March 2026

The decision moves into the model. The purchase still happens in the store, but by then the shortlist is set.

Open with behavior: users increasingly decide with a language model, execution still happens in the store. 43 percent already research products with an assistant, and those visitors convert better once they arrive.

What this means for our clients

Discovery is collapsing into the answer.

The consequence for our clients: ten links became one answer. A brand that is not in the answer does not exist for that shortlist.

RECORDED LIVE RUNS · OUR MEASUREMENT INFRASTRUCTURE · TEMP 0

“What is the best online shop for bikes and bike parts?”

GPT-5.5
Claude Sonnet 4.6
Gemini 2.5 Flash

R · replay

Same prompt, same instant, three answers. Only one names us. Press R to replay the three panels live, safe to repeat.

Where the traffic goes typical online retailer · stylized mix from public benchmarks

Then · before AI answers
Organic search 38%Direct 26%Paid 18%

Organic search 38% · Direct 26% · Paid 18% · Email 6% · Social 5% · Referral 7%

Today
Organic search 28%Direct 27%Paid 18%

AI 2% · decided in the answer 8% · Organic search 28% · Direct 27% · Paid 18% · Email 6% · Social 5% · Referral 6%

Retailer shopsExecution traffic largely persists: people still check out somewhere.
Brand sitesAdvisory traffic is exactly what moves into the model. Manufacturer sites take the bigger hit.

Pew 2025: clicks drop 15% → 8% under AI summaries · Gartner: −25% search volume by 2026 · Adobe: AI referrals small but +393% YoY

The AI channel itself is tiny as a source, low single digits. Its real effect is the slice it removes: decisions that end inside the answer and never become a click. Organic pays the bill, and brand sites more than retailer shops, because advisory queries are their traffic.

Why this matters for Y1: three perspectives

ImplementationAre our clients' shops visible where buying decisions now happen? (GEO Scan)
ConsultingWhen users ask the model for advice: which product ends up recommended? (Product Visibility)
SalesThe same mechanics decide whether Y1 shows up when someone asks a model for a partner. Our visibility counts too.

Land the three angles: GEO for client shops, product visibility inside model advisory, and Y1's own visibility in the same answers.

An answer is a pipeline, not a page.

Transition into frameworks. An answer is produced by a pipeline, so optimization has to address a stage, not 'the AI'.

The Four Stages

keys 1-4 or click · select stage

Step through the stages with keys 1 to 4. Each stage has its own tactics and its own failure modes.

The GEO periodic table 33 tactics · geo.y1.studio · interactive, click and scroll inside

Live embed, no key needed. Click an element to open a tactic, or just scroll and narrate.

Unbranded vs Branded

Unbranded · the discovery KPI

“What is the best online shop for bike parts?”

The model names you, or it doesn't. That rate is real visibility.

Branded · a token echo

“Is BIKE24 a good shop?”

The answer contains your name by construction. It measures recall, not discovery.

Our edge:we report them separately. Most tools don't.

Core methodological point: unbranded prompts measure discovery, branded prompts mostly echo the brand token back. We report them separately, most tools do not.

The model decides, not your website alone.

0

welt.de citations on ChatGPT (Agora study, 4,570 citations)

0

welt.de citations on Claude (same study, same prompts)

Axel Springer × OpenAI licensing deal, Dec 2023. You just saw the 33 things you can control. This number is decided elsewhere: next slide.

On-table vs off-table: two levels decide

Off-table · the layer above your website Licensing deals · retrieval backends · syndication decides how often you actually appear, and in which model
On-table · your website and content the 33 periodic-table tactics: they decide whether you can appear at all

welt.de, resolved: 264 citations on ChatGPT vs 0 on Claude is not a content problem. It is an off-table licensing deal (Axel Springer × OpenAI).

Resolve the welt.de puzzle from the previous slide: on-table work explains eligibility, off-table deals explain amplitude.

We don’t audit opinions. We run measurements.

GEO Scan · brandsIs the brand visible in AI answers, and does the domain comply with our GEO framework? Audit against the 33 tactics, plus live measurement per model and language.
Product Visibility · productsWhich product does a model recommend in which usage context? Product-level, context-specific measurement: the research instrument.

0

GEO scans (brand level)

0

Product Visibility scans (Stihl, HiPP, Scott Sports)

0

≈ measured model answers

The two instruments in one view: GEO Scan for brand visibility and framework compliance, Product Visibility for product-level context. Everything from here is one of the two.

How a GEO scan works Instrument 1 · brand level

DE EN FR IT ES NL GPT-5.5 Claude Sonnet 4.6 Gemini 2.5 Flash

prompts × models × languages × temp 0 → share of voice

$0

total cost of the five complete scan batches (1,915 answers)

0.7 ct

per measured answer

Instrument 1 mechanics: same prompts, three models, six languages, temperature 0. Cheap enough to rerun anytime.

A real GEO scan report: Lusini reports.y1.studio · interactive, scroll inside

Live embed, no key needed. Scroll within the frame to the section you want, this is the real client-facing report.

Models do not agree: Lusini share of voice by model

Share of voice by model, %

LUSINI on Anthropic: 0%. Amazon on Anthropic: 49%.

Punchline of instrument 1: models disagree massively. LUSINI at zero on Anthropic while Amazon takes 49 percent there.

Product Visibility: contextual product intelligence (Stihl pilot)

240 prompts

24 contexts

2,400 runs

12,827 mentions

MS 261 C-M 44.7%
MSA 300 29.2%
HSA 56 24.0%
RE 140 0.2%
+88 pp vs competitors · semi-professional construction context −48 pp · beginner first-purchase context (Husqvarna 120 takes 96%)

87% of all mentions were products outside the pilot catalog. The model recommends a wider market than any brief.

Instrument 2, product level: 24 usage contexts, and 87 percent of all mentions were products nobody briefed.

A real Product Visibility report: Stihl the client deliverable · interactive, scroll inside

Scroll the actual Stihl report: mention rates, ranks, context heatmap, head-to-head against Husqvarna and Makita. This is what the deep dive delivers.

Seven brands later, the patterns repeat.

Act break. From here: what seven brands taught us, pattern by pattern.

Blocked at the Door

verpackgoParent brand named in 94% of answers, the shop itself in just 2%, organically. WAF 403 · GPTBot
LusiniClaudeBot passes; GPTBot and PerplexityBot get challenged: 50,000 items missing from ChatGPT. cf-mitigated: challenge
Lusini / WikidataThe machine record still says “E.M. Group.” Q17539735

Crawler-access failures: verpackgo and Lusini each lose a whole answer surface without noticing.

Present but Wrong, or Absent Where It Counts

Stihl ALLPRONamed in 82% of answers, described correctly in 0%. merged with old AP system
MelittaIts best advisory content is blocked by robots.txt. robots.txt Disallow
Cybex88% share of voice across 10 languages, 49% in English, 0/6 in US buying contexts. en cliff

Content and entity failures: ALLPRO is named but misdescribed, Melitta blocks its best content, Cybex collapses in English.

The 55% Question

Which shop holds 55% share of voice in bike retail?

G, click, or → · reveal

Wiggle.

0%

2023 insolvency, brand relaunched smaller under Frasers Group: 90% share of voice on Anthropic

Training data is not reality.

Ask the room to guess out loud before revealing. Press G, or click anywhere on the slide, to reveal Wiggle and the 55% figure. Precise history: administration Oct 2023, brand and IP bought by Frasers Group March 2024 for under 10M pounds, relaunched smaller. The 90% on Anthropic reflects the pre-collapse Wiggle: training data lags reality by years.

Three instruments: when to use which

GEO Tactics · frameworkHow can a brand work on its visibility, where, per model and language?
AI Visibility Scan · based on GEO TacticsHow visible is a brand with its sites and shops, and how is that visibility developing over time?
Product Visibility · deep diveHow visible are specific products in specific contexts, and how do they compare against competitors?

The stack: GEO Tactics is the framework, the Scan measures a brand against it over time, Product Visibility drills into specific products and their competitors.

Everything is open.

AI Visibility Scan

ai-visibility.y1.studio

GEO Periodic Table

geo.y1.studio

GEO LinkedIn article

linkedin.com/pulse/word-machine

Every number in this deck traces to a repo artifact. For interpretation, measurement and client projects: start with me.

Leave this up and let people scan. Everything is in the repo and the reports.

Thank you!

Leave this up during Q&A. Contact: Tobias Becker, t@bckr.bio.