How AI answers are changing the search channel, what the opportunity is, and what we can and cannot measure. Written for executives who want straight answers before anyone sells them a tool.
20 slides + 4 appendix · Interactive · Sources in the appendix
The deck follows the order in which executives usually ask their questions: why this matters, how big it is, how an AI answer is actually formed, and how anyone could know whether it is working.
Most slides respond when you click them: you can build an AI answer one piece at a time, or switch the market data between months. The text under each slide carries the full explanation. Click into the deck and use the arrow keys to move through it.
Every market number comes from public Similarweb posts and is listed in the appendix with its date and scope. Examples use a neutral category, noise cancelling headphones, and brands labeled A, B and C.
TNC Visibility - An explainer for executives
AI visibility, explained
How AI answers are changing the search channel, what the opportunity is, and what we can and cannot measure.
The question
"What are the best noise cancelling headphones for a long flight?"
Review site: travel headphones, tested
Retailer: noise cancelling headphones
Forum thread: headphones for long flights
Brand A: product page
Guide: what to pack for a long-haul flight
Brand A has the best battery for long flights[1]. Brand B has the strongest noise cancelling, at a higher price[2]. Brand C is good value but weaker on calls[3].
Illustrative example. Brands are labeled, not real.
Dennis Goedegebuure, 2026
AI visibility, explained - TNC VisibilityExample is illustrative.1 / 24
This deck answers four questions in order: why AI answers matter to a business, how big the shift is, how an AI answer actually gets formed, and how anyone can track it honestly. It ends with what to do with that understanding. It does not end with a sales plan.
Opening - The executive basics
Why executives are asking about this now
Customers
They ask AI assistants for recommendations, not just facts.
Boardroom
Boards and competitors raise it, with numbers nobody can check.
Platforms
Ads and shopping are being added inside answers.
Own team
Asked: why do we not show up there? No shared answer yet.
The pressure is real before anyone knows what to buy.
What is missing in most companies: a shared, honest answer.
This deck is that answer, including where measurement does not exist.
AI visibility, explained - TNC VisibilityFraming slide. No external data.2 / 24
Notice what is missing from this list: a vendor, a tool and a tactic. The pressure is real before anyone knows what to buy. This deck is the shared answer, including the parts where the honest answer is that measurement does not exist yet.
Opening - The executive basics
The shortlist moved inside the answer
Search
Question
Results page
Five site visits
Shortlist of 3
Purchase
AI answer
Question
One answer, shortlist of 3 with reasons
A few clicks out to sources
Purchase
In search, the customer built the shortlist by visiting sites.
In an answer, the engine builds it and gives reasons.
Ranked but not mentioned is a new kind of invisible.
AI visibility, explained - TNC VisibilityModel: TNC Visibility.3 / 24
This is the whole importance argument in one move of the shortlist. Everything later in the deck is either evidence for this, an explanation of how the machine builds that shortlist, or an honest account of how hard it is to watch it happen.
Opening - The executive basics
How big is it: attention already moved
Month on month change shown next to each value.
Monthly visits, worldwide, September 2026. Universe: all websites, top 10,000 domains.
14.4 times
google.com visits versus chatgpt.com. Google is not disappearing.
Up 5.00%
ChatGPT in September 2026, its second highest month ever. It now sits ahead of Bing, Reddit and TikTok on visits.
Cited twice over
YouTube (No. 2) and Reddit (No. 7) matter again as sources answers cite.
AI visibility, explained - TNC VisibilitySource: Similarweb, Top 10 websites, September 2026.4 / 24
Two cautions I will keep repeating. These are worldwide website visits, not a brand's market and not AI answer impressions, which nobody publishes. And size is not the argument for panic. Google is still 14.4 times ChatGPT. The argument is that a second, large discovery surface now exists next to the first one.
Opening - The executive basics
Why a brand should care
Illustrative answer
"What are the best noise cancelling headphones for a long flight?"
Brand A: best battery for long flights.
Brand B: strongest noise cancelling, at a higher price.
Brand C: good value, weaker on calls.
Sources cited: two review sites, one retailer, one forum thread.
Your brand: not mentioned. No story at all.
Select a card to see where it sits in the answer.
AI visibility, explained - TNC VisibilityExample: TNC Visibility illustration. Real answers vary by user, session and run.5 / 24
The last line is why this is a workable problem and not fate. The same public record that built the model's view keeps being read, in training over time and in lookup every day. Slide 14 returns to what a brand controls in that record.
Section A - How AI forms an opinion
Search ranks pages. AI predicts answers.
Search
An index, ranked
Pages are stored on an index shelf and ranked for a query: positions 1, 2, 3 and so on. A position can be checked by anyone.
AI
Prediction, piece by piece
Illustrative weights, not taken from any real model. Bar width shows how likely each next piece is.
No index. No ranking. Each piece is a token (roughly a word or part of a word), and a brand's standing in a model is the probability of being predicted, nothing else.
AI visibility, explained - TNC VisibilityModel description: standard language model behavior.6 / 24
This is how a language model writes: token probability built from training data. Hold on to the practical meaning. There is no position 3 in an answer to defend. There is a tendency of the model to reach for certain brands and certain claims when a question is asked. All work in this channel is about shifting tendencies, not slots.
Section A - How AI forms an opinion
Memory and lookup: the two sources of an answer
Try a question type:
Most answers draw on both.
Memory (slow)
Patterns absorbed in training: category, brands, claims, rivals. Decides most of what is said about settled topics. Changes over months to years, as the record changes and models retrain.
Lookup (fast)
Live search and feeds at answer time: results, prices, news, product data. Covers current, specific and weakly learned facts. Can change in weeks when sources change.
One answer
AI visibility, explained - TNC VisibilityModel: TNC Visibility, October 2026.7 / 24
Vendors call memory parametric. The plain word is fine here. The strategic point is the speeds: a brand that fixes its sources and feeds can see lookup driven answers move quickly, while the memory layer only follows after the public record has been consistent for a long time. Anyone promising fast memory change is selling something.
Section A - How AI forms an opinion
Query fan out: one question becomes many searches
Filter by source type:
The prompt
"Best noise cancelling headphones for long flights?"
One question is all the customer sees being asked.
The fan out: 8 sub questions (illustrative)Mostly invisible to the customer and to brands.
The answer
One written answer, built from the sub question results.
The real competition happens in the middle column, out of sight.
AI visibility, explained - TNC VisibilityDiagram: TNC Visibility. Sub questions are illustrative.8 / 24
Fan out is why a brand can lose an answer it never saw being built. The customer asked one question about headphones for a flight. The model quietly asked eight of its own, and the brands that showed up in those eight answers wrote the ninth, the one the customer read. Any tracking method that only runs the headline prompt is watching the tip of this.
Section A - How AI forms an opinion
Every answer is individual, so one test proves nothing
One prompt: "What are the best noise cancelling headphones?" Pick a condition to see what comes back.
Shortlist returned
Same prompt. Different answers. Past activity, settings, location, language and sampling between runs all feed in as well.
AI visibility, explained - TNC VisibilityBehavior: personalization and memory features of consumer AI products, October 2026. Orders shown are illustrative.9 / 24
So when someone senior asks the assistant a question at home and the brand is missing, that is one answer, for one account, with one history, on one model version. It may match what customers see or may not. Slide 18 shows what measurement looks like once we accept this slide as true: samples, repeats and averages, never one screenshot.
Section A - How AI forms an opinion
A brand, to a model, is a bundle of associations
Strong lines
Category, audience, strengths: built by consistent public evidence.
Thin and dashed lines
Gaps do not stay empty. Reviewers, forums and rivals fill them, and the model repeats what it finds most often.
AI visibility, explained - TNC VisibilityDiagram: TNC Visibility illustration. Line weight is illustrative.10 / 24
Gaps do not stay empty. Reviewers, forums and rivals fill them, and the model repeats what it finds most often and most consistently. That is the bridge to section C: earning visibility is record building, done where the model and its lookups actually read.
Section B - Market, players and demand
Three players hold AI attention, and Google holds both games
Compare Oct 2025 with:
Oct 2025Sep 2026
Share of website traffic to Gen AI sites only. Worldwide.
A year ago one surface was the market
ChatGPT 71.8%, Gemini 13.6%, Claude 2.3%. Now it takes three, plus Google AI answers on top of classic search. And Google owns Gemini while running the No. 1 search site: 85.15B visits.
In the whole web: top 10 websites, September 2026
Monthly visits. Universe: all websites, top 10,000 domains. A different measure from the Gen AI share chart on the left. Do not mix the two.
AI visibility, explained - TNC VisibilitySource: Similarweb, Gen AI share update and Top 10 websites, September 2026.11 / 24
Do not read the ChatGPT line as decline. Its visits hit their second highest month ever in the same period. Share fell because the market grew and split. For planning, the count of surfaces to earn and to track is the takeaway: ChatGPT, Gemini, Claude, plus Google AI answers on top of classic search.
Section B - Market, players and demand
Estimating AI demand: a method, not a measurement
Step 1
Measure
Classic search demand, from keyword research. The one demand signal that is measured.
Step 2
Cluster
Group keywords into customer questions and intents, not single words.
Step 3
Allocate
Give AI platforms a share of that demand, using their share of attention. This is the assumption dial.
Step 4
Range
Present a range with the assumption printed, and show how it moves if the dial is halved or doubled.
No platform publishes prompt volume, impressions or category splits. Any AI demand figure is search demand reallocated by stated assumptions. Useful for prioritization. Useless as a reported actual.
Allocation anchors: Gen AI traffic share by player (slide 11) and AI site visit scale versus search (slide 4).
AI visibility, explained - TNC VisibilityMethod: TNC Visibility. Anchors: Similarweb, September 2026.12 / 24
Two allocation anchors exist today: the Gen AI traffic shares on the previous slide, and the visit scale of AI sites versus search sites from the Top 10 table. Neither is prompt volume by category, and this slide must never be quoted as if it were. Its use is ranking opportunities against each other under one shared set of assumptions. Appendix A3 shows the arithmetic on illustrative numbers.
Section C - What visibility means
What good visibility looks like
Right column: illustrative snippets from the headphones example.
AI visibility, explained - TNC VisibilityFramework: TNC Visibility.13 / 24
Each line is also a failure mode in disguise: named last, named for a rival's reasons, strong on one platform and absent on the others, cited to a source nobody can affect, or visible once in a screenshot and gone the next week. The last line is the one that separates visibility from luck, and it needs the tracking method from section D to judge.
Section C - What visibility means
SEO builds the record. AI answers read the record.
Outcome 1
Search rankings and clicks
The SEO results leaders already report on.
Outcome 2
Mentions and citations in AI answers
The new surface, fed by the same record.
Both rise from one base
Product data
Clear data on products, prices, uses and audiences.
Answering pages
Pages that answer customer questions in plain language.
Third party evidence
Reviews, guides, video and forums, where engines read.
Consistent profiles
The same facts on the site, at retailers, in profiles.
Trap: Cutting SEO to fund AI tactics starves the very sources AI answers are built from.
AI visibility, explained - TNC VisibilityModel: TNC Visibility.14 / 24
Both pipes from slide 7 drink from this record. Lookup reads it live. Training absorbs it over time. That is the entire relation between the channels: one budget line builds the evidence, the other is where the evidence increasingly gets used. Judge any proposed AI tactic by one question: what does it add to the record, and where will engines read it.
Section C - What visibility means
AI plays a broadcast role in the journey
Step through the three stages.
Research stage
The answer plants the seed
At research time, an AI answer names brands and attaches reasons. Like a TV ad, it works on familiarity and trust, often below conscious recall.
Later, elsewhere
The sale happens on another channel
Days later the same customer types the brand name into search, walks into a store, or buys at a retailer. No click ties the purchase back to the answer.
What moves
Signals on other channels
Branded search volume rises. Retailer visits and sales shift. Conversion rates on other channels improve, because more arrivals already trust the brand.
AI visibility explains part of the journey, not all of it. That is normal for a broadcast medium.
Part of playing in this arena is judgment: being absent while rivals are named is also a decision.
Measure this layer at brand level: awareness and brand health studies, slide 18.
AI visibility, explained - TNC VisibilityModel: TNC Visibility. Measurement: slides 16 to 19.15 / 24
TV advertising is the comparison executives already understand. Nobody asks a TV spot for its click through rate. Its job is to be remembered at the moment of choice, on a different channel, later. AI answers at the research stage work the same way, which is why the tracking section adds brand health measurement next to answer sampling.
Section D - Tracking and reporting
Why AI visibility is hard to track: five structural reasons
1. No fixed output - answers vary by user, session and runSearch: one results page to check
2. Platforms publish almost nothing - no console for organic answersSearch: platform reported impressions
3. The fan out is mostly hidden - not shown in full by default, though parts can be surfaced if you know howSearch: the query is the query
4. Clicks are scarce - referral data from answers is thinSearch: clicks carry source data
5. Models change silently - all brands can move at onceSearch: updates are announced and studied
AI visibility, explained - TNC VisibilityAnalysis: TNC Visibility, October 2026.16 / 24
Compare this with search, where a ranking can be checked by anyone, impressions are reported by the platform itself, and clicks land in analytics with sources attached. None of those three legs exists here in a dependable form. The next three slides name exactly what is missing, then show the working method the industry is converging on, assumptions included. One correction on the fan out row: parts of the fan out can be made visible with the right methods and tools; what platforms do not provide is the complete official record.
Section D - Tracking and reporting
The metrics we are missing
AI answer reporting, as platforms publish it today.
AI visibility, explained - TNC VisibilityStatus as of October 2026: TNC Visibility review of platform reporting.17 / 24
Read the list as a set of things nobody has, including the platforms' own partners and every vendor in this market. If a number in a future report looks exactly like one of these, the right response is one question: where did that come from. Slide 19 turns that instinct into five standard questions.
Section D - Tracking and reporting
A workable view: samples, proxies and stated assumptions
Fixed question panel
Customer questions, chosen to represent demand, then frozen.
Repeat runs
Same panel, run again and again under named conditions.
Average and trend
Share of answer over time. Single runs are not reported.
Report with assumptions
The register travels with the numbers, every time.
Brand health studies
For the broadcast effect (slide 15): awareness, consideration and preference, measured in studies over quarters. Branded search volume as a supporting signal.
Illustrative shape only. Trends are reported, single runs are not.
Assumptions register, example entries
Panel questions represent category demand. Runs use fresh sessions unless stated. Model version is noted per run. Citation tracking covers the observable part of the evidence.
AI visibility, explained - TNC VisibilityMethod: TNC Visibility.18 / 24
Three disciplines make this honest. The question set is fixed and chosen to represent customers, not to flatter the brand. Runs are repeated under named conditions, accounts, memory states and models noted. And the assumptions register travels with the numbers, so a reader can see which conclusions depend on which hypothesis. A trend with its assumptions showing beats a precise looking figure with its assumptions hidden, every time.
Section D - Tracking and reporting
Five questions to ask any AI visibility report
Hold a report up against this list. Tick each question it answers well.
Apply the test to agencies, tool vendors and internal teams alike. Vendor models, for example neat splits between training and lookup influence, are claims to test, not settled facts.
AI visibility, explained - TNC VisibilityFramework: TNC Visibility.19 / 24
A good report answers all five without being asked twice. Apply the test symmetrically: to agencies, to tool vendors, and to internal teams, including us. Claims heard in this market recently, for example neat splits between training influence and lookup influence, should be presented as a vendor's model to test, not as settled fact.
Close
What this means
1. The shortlist moved inside the answer.
2. A few platforms hold the attention, Google among them twice.
3. A brand's public record is what those platforms read.
4. Measurement is partial: trends and stated assumptions.
The follow on
Apply this framework to one brand:
One brand. Its category questions. Its record, read like a model reads it.
That is a separate strategy deck and a separate conversation.
AI visibility, explained - TNC VisibilitySummary: TNC Visibility.20 / 24
That last line is an offer, not a slide in this deck. Applying the framework means mapping one brand's category questions, reading its public record the way a model would, and setting up the sampling method from slide 18. That work exists as a separate strategy deck and a separate conversation.
Appendix
Sources
Gen AI share
Gen AI website traffic share, last 12 months as of September 2026
Universe: Gen AI sites only. Worldwide. Similarweb post, October 12, 2026.
Every chart in this deck is rebuilt from the tables in Appendix A2.
AI visibility, explained - TNC VisibilitySources: Similarweb posts on X, October 2026.A1 / 24
Every external number in the deck traces to one of these three posts. Each one states its universe; the Gen AI share and the all-web ranking are different measures and are never mixed.
Appendix
Full data tables
Table A: Gen AI share, worldwide (Gen AI sites only)
Table B: Top 10 websites, September 2026 (top 10,000 domains)
AI visibility, explained - TNC VisibilitySources: Similarweb, September 2026 data.A2 / 24
The exact numbers behind slides 4 and 11, reproduced without rounding changes.
Appendix
Worked estimation example (illustration)
Illustration only. Round invented inputs, used to show the arithmetic. Not market data.
Step 1 and 2, invented
Invented input. Real work uses measured keyword research.
Step 3, invented dial
5%10%20%
AI allocation settings. Invented settings.
Step 4, modeled range
Real work replaces the first input with measured keyword research and defends the dial setting with platform share data. The output stays a range.
This is the only place in the deck where invented numbers appear, and only under the banner. Change the inputs to show the room how much the output depends on the dial. That dependence is the point.
Appendix
Assumptions register, template
If an assumption falls, the conclusions in its row fall with it. That is the point of writing them down.
Assumption
Why it is needed
What depends on it
How it could be tested
The question panel represents real category demand
No prompt volume exists
Everything reported as share
Compare panel topics with search keyword clusters
Fresh sessions approximate a typical user
Account memory varies per user
Absolute mention rates
Run a parallel panel on accounts with history
Model versions are comparable within a month
Providers change models silently
Month on month movement
Re-run after a known change, flag the series break
Visible citations represent the evidence used
The fan out is mostly hidden
Source strategy
Compare citations with observable retrieval results
AI visibility, explained - TNC VisibilityTemplate: TNC Visibility.A4 / 24
This is the document that travels with every AI visibility report. If a report arrives without one, ask for it before reading the numbers.
The full explanation
Click into the deck, then use the arrow keys to move between slides. N shows or hides the full explanation.
Apply it to one brand.
One brand, its category questions, and its public record read the way a model reads it. That is a separate conversation.
Cookies, your call. We measure site statistics with Google Analytics, but only if you accept. No marketing cookies, no surprises. See the cookie policy.