The answers are hardening.

You have been seeing it.

A note from one founder to another

I am writing this as one of you.

The advice in the pages that follow is what I am applying to my own practice in 2026. Whatever does not survive that test does not appear here.

This is the second edition. I published the first earlier this year, and enough has moved since then that reprinting it would have been dishonest. Every figure below is current to August 2026, and where I have corrected myself, I say so rather than editing the number and hoping nobody checks.

Two things I will not do. I will not pretend the picture is more certain than it is, and I will not pretend the time available is longer than it is. The next fourteen months matter more than the fourteen that follow.

Alexandre Olive

What changed, in plain words.

Three terms to have straight before going further.

Four lines tell the whole arc

Stanford HAI, AI Index 2025 and 2026

Why the next fourteen months matter.

Four shifts already in motion.

Shift one

Your client is already using it.

Bain and Altagamma, in the June 2026 update to their worldwide luxury study, put it plainly: roughly half of luxury consumers already use AI somewhere in their buying journey, and nearly all of them intend to carry on. One in four use it to discover brands and products. Two in three use it to compare. Bain names the disrupted funnel as one of the four forces reshaping the market this year.

The client who walks into your salon on Tuesday morning has spent the weekend asking an engine what to expect, what to ask, what the alternatives are, and whether you are the right house for the piece. Whatever the engine said is the brief the client is now holding.

Federica Levato, who co-authors the study, frames the test for the industry as whether brands are building the relevance to be “surfaced and chosen” when the client returns. That is a strategist’s sentence, and it is also correct.

Bain & Company with Altagamma, Luxury Goods Worldwide Market Study, spring update, 25 June 2026

An exercise, before you read further

They are also asking the AI about you.

Open ChatGPT now. Type your house’s name. Then ask: who are the leading houses in my category, in my city, at my price.

Three possibilities. You are not in the answer. You are in the answer but described inaccurately. You are in the answer and described in a way that makes you sound like the three houses two streets down.

Each of these is a different problem with a different fix, and the most cost-effective hour of strategic work a founder can do in any given quarter.

Shift two

What your private client wants has not changed.

Three things, above all. Managed expectations, a relationship with specific people, and their time respected.

AI delivers all three when it is put to work inside a house that knows its voice and its boundaries. It destroys all three when it is put to work without either.

Shift three

The meaning of the purchase is moving.

Bain’s 2026 reading is that the meaning of luxury is shifting from social validation toward self-actualisation, from being admired toward living well. Consumer appetite for experiences is outgrowing tangible goods by one and a half times. Half of luxury shoppers now check the secondhand market before buying anything new.

And this, which anyone running an independent house should read twice: in watches, connoisseurship is overtaking hype, with collectors rewarding craftsmanship and rarity.

The industry’s own analysts are now describing the conditions independents have been building for over two decades. The demand is arriving. The open question is whether the house is findable when it does.

Bain & Company with Altagamma, June 2026

Shift four

Smallness is now an advantage.

Eurostat’s latest reference year puts AI use at 20.0% of EU enterprises with ten or more employees, up from 13.5% twelve months earlier. Among large enterprises the figure is 55%. Among small firms it is closer to one in six.

That gap was produced by a capability constraint that no longer holds. LVMH spends years aligning a portfolio of maisons. A ten-person house decides on Monday and ships by Friday. A project that required five million euros and a full engineering team in 2022 now assembles for forty to a hundred and fifty thousand euros, in three to six months, with the right architecture. That band covers a full build of several instruments.

The small house that moves now closes the gap before the large one has time to compound.

Eurostat, use of AI in enterprises, 2025 reference year, published December 2025

The objection worth addressing

If the technology gets cheaper every six months, why now?

The technology gets cheaper. Four things do not.

• Your corpus.The voice, archive, client history, and internal knowledge that calibrates these tools to your house specifically. Eighteen months of corpus work in 2026 produces an instrument the house starting in 2028 cannot replicate without going back in time.
• Your position in the answers.Authority in AI search compounds the way search authority compounded between 2008 and 2015.
• Your team’s fluency.Forty to sixty hours of practice per person, on the practice’s own count, does not get faster because the tools improve.
• And a fourth.Which I could not have written in the first edition because the measurement did not exist.

What the engines are saying about independent houses.

Being cited is not the same as being recommended.

Almost everyone selling visibility work in 2026 is selling citations: the small links underneath the answer. The same study found that only 21% of the most-cited domains in a category were also the most-mentioned brand, and that the two signals correlate slightly negatively. Earlier work by the same author found that roughly three in four readers pick the brand named in the answer text as their choice.

Your client reads the answer. Nobody reads the footnotes. A house can be cited constantly and recommended never, and most of the industry is chasing the footnotes.

A pilot, earlier this year

What I found when I asked the engines about watchmaking.

I took the questions a wealthy buyer asks, the ones that come before a boutique visit rather than after, and put them to ChatGPT, Gemini and Perplexity. Then I read what came back, with twenty years in the industry behind the reading rather than a scoring rubric.

Three things held across all three engines.

• 01

  All three deflected the buyer away from retail using near-identical depreciation figures, none of them sourced to anything. The same number, phrased three ways, presented as fact.

• 02

  A certified horological first was stated in full by one engine and softened by another, because the second had drawn on forum posts rather than the maison’s own record. A four-year development, adjudicated by a message board.

• 03

  One independent maison appeared in none of nine intent-based buying questions, while the same four or five rivals filled every single answer. The house did not place low in those answers. It was not in them.

None of this is the engine behaving badly. It is the engine doing exactly what it was built to do with the material available to it. The maison had a hundred years of craft and almost nothing a machine could read.

I did not set out to build a visibility practice. I set out to read what the machines were saying about houses I have known for twenty years, and I did not like the answers.

Running this month

The Report.

How AI engines describe the luxury industry

That pilot has become. The Luxury AI Report is published by ELIA Atelier and measures how AI answer engines describe luxury houses. Edition one, published 27 August 2026, records 17,820 answers across 46 Swiss watch houses, nine buyer questions, four engines and three markets.

Through August 2026 those forty-six houses are read across ChatGPT, Gemini, Claude and Perplexity, against the buying questions their clients ask, in English. Edition one publishes at 07:00 CET.

The large visibility studies published this year cover software, retail and consumer categories, where the samples are easy and the buyers are obvious. The independents, the segment with the most craft per employee and the least machine-readable record, have been measured by nobody. That is the gap the Report fills, and the timing is not incidental: Swiss watch exports came off two consecutive declining years before June 2026 returned 11.2% growth, and a segment recovering into a rebuilt buying journey has more riding on the answer layer than one in a flat market.

I am publishing it because the independents have the most to lose from an answer layer that defaults to the four or five names it has read about most, and because a house cannot correct a picture it has never seen.

Federation of the Swiss Watch Industry, monthly export statistics, June 2026

On whether this holds beyond watchmaking

I expect it does. Nothing in what the pilot surfaced is specific to watches. The engines were not penalising horology; they were reaching for the material available to them and finding forums, resale listings and press aggregation where the house’s own record should have been. A jewellery house, a perfumer, a leather atelier and an architectural practice all share that shape: deep craft, a founder who carries the story in person, a thin published record, and a category where four or five large names dominate everything written online.

So I expect the pattern to repeat, and I am not going to claim it until it has been measured. The cohort is watchmaking, and I will not claim beyond it. The reason to say this plainly rather than extend the finding without saying so is that the whole argument of this paper is about what happens when unsourced claims get repeated until they harden into answers. I would rather not add to the pile.

Jewellery and perfumery are next in the programme. If you run a house in either category and would like your answers read as part of that cohort, write to me.

What the work consists of

Three parts, in order, and the order matters.

• First, the reading.What the engines currently say about the house, across the questions its buyers ask, in the languages its buyers use. Most founders have never seen this, and the first sitting is usually uncomfortable.
• Second, the corpus.The house’s own record, written so that a machine can read it without flattening it. Not more content. The right content, in the house’s voice, saying the specific things only this house can say.
• Third, the third-party record.Research from the University of Toronto found that the overwhelming majority of AI answers cite third-party material rather than the brand’s own site, and that brands are several times more likely to be surfaced through someone else’s pages than their own. Your website is necessary. It has never been sufficient, and it is now less sufficient than it has ever been.

The goal is not traffic. The goal is that when a serious buyer asks a serious question, your house is in the answer, described in terms you would have chosen. Findable, and unmistakable.

Three things nobody tells you.

One

Most AI projects fail before they produce anything.

MIT’s NANDA initiative reviewed around three hundred publicly disclosed deployments and found that roughly 5% produced measurable impact on the profit and loss. The cause named in the report was not model quality. It was what the authors call the learning gap: organisations unable to fold the tools into how the work is done.

I use that figure with a caveat, because the study has been quoted far more often than it has been read and its method is thinner than the headline implies. I use it anyway, because it matches what I see in the field. The houses that fail start three things at once. The house that starts with one well-scoped instrument and finishes it before beginning the next has something working by month nine.

MIT NANDA, The GenAI Divide: State of AI in Business, 2025

Two

The human layer is the biggest risk.

McKinsey has been counting failed transformations since long before AI, and the number has held at roughly 70% across every technology wave of the last thirty years. The cause has almost never been the technology.

Inside an independent house, the cohort most resistant is almost always the administrative staff. The sales coordinator, the proposal writer, the marketing assistant, the bookkeeper, the office manager. The roles where the work is most addressable by the new tools, and the roles where the fear of replacement is most rational.

The moment the founder says the word AI in a meeting, the team’s trust takes a hit before any tool arrives. Whatever the founder says next has to address the fear directly, or the implementation is already losing ground in its first week.

What the founder is seeing

Picture the senior salesperson who has spent eighteen years writing letters to her clients in her own voice, asked to let a machine draft them.

The atelier head whose juniors used to come to him for research that now takes seconds.

The marketing coordinator who learns that the part of her job she trained for is the part the machine does best.

None of these people is resisting the technology. They are mourning a part of themselves that the work used to hold.

Training alone does not address it. What does is the founder naming what is being given up, in the same room where the new tools are being introduced, before anyone is asked to celebrate them.

Three

What makes you defensible is not the technology.

The technology is rented. Anyone can buy your ChatGPT subscription.

Your know-how, archives, reputation, client relationships, aesthetic and judgment are what make the house irreplaceable. None of those is reproducible by a competitor with the same subscription.

AI introduced inside a house that has named its voice and its boundaries encodes those assets and protects them across every surface where the house speaks. Introduced without that naming, it turns the same assets into output indistinguishable from everyone else in the category.

Where to start. Two moves.

You do not need more than this to begin. The house that does these two well in the next six months has its foundation in place for everything that follows.

The move that makes you singular

One signature instrument.

Every serious house has had at least one signature instrument over its history. A bottle, a complication, a scent formula, a collection, an experience. The thing that did more for the house than a hundred campaigns, because it carried the story, the product and the proof inside one object the client could hold.

In this decade, the equivalent is a digital product built on these tools and shaped to the house’s craft. Chatbots do not qualify. Something the house builds once, that no other house in its category can replicate, and that tells the house something about its clients it did not previously know.

Where most houses find it

• i
  A client memory layer.

  Holding everything the house knows about each client across every interaction, queryable by the founder, with selective access for senior staff on agreed topics.

• ii
  An automated proposal system.

  Drafting in the founder’s voice, drawing on the archive and case history, producing a personalised document the senior salesperson refines in an hour rather than a day.

Build cost for a single instrument of this kind sits at the lower end of that band, forty to eighty thousand euros for a house in this segment. Runs three to five years. Payback within six to twelve months in almost every case where the team uses the instrument well.

A reframe

Where this belongs in the budget.

A marketing campaign produces a six-month run of impressions and goes back to zero. A trade-show booth produces three days of conversations and is taken down. Office rent resets each year. None of these compounds.

The instruments described here produce three things that compound from the day they go live. Productivity, hours returned to the team across every working week, for three to five years. Output quality, the proposal that arrives in the client’s hands with the house’s voice intact, the relevant precedents drawn from the archive, the personalisation that closes the work. Defensibility, corpus, position in the answers, team fluency, all building inside the house and unavailable to competitors who start later.

The founder who books this as an expense will underbudget it, undermeasure it, and undermanage it.

Four things to avoid

• i

  Do not hire a Chief AI Officer under a hundred employees. Hire one experienced person, fractional, on a six-month engagement.

• ii

  The shiny is not for you yet, and 2026 has already provided the worked example. OpenAI launched in-chat checkout in September 2025 with a great deal of noise about the end of the shopping page, and retired it in March 2026 after roughly thirty merchants had integrated. The protocol survives. The consumer product did not. Meanwhile the discovery layer, the part nobody demonstrated on a stage, is where your client already is. Digital twins, virtual showrooms, AI influencers: say not yet without saying never.

• iii

  The founder cannot be the only person who understands the new tools. Within six months of starting, a second person must know how the work was built and why.

• iv

  Governance is not optional, and the timetable changed under everyone’s feet this summer. The EU’s Digital Omnibus, endorsed by Parliament on 16 June and approved by the Council on 29 June 2026, deferred the high-risk obligations of the AI Act to December 2027 and August 2028. It did not defer Article 50. The transparency duties, including telling a person when they are speaking to a machine and marking synthetic content, applied from 2 August 2026. Fines for the prohibited practices reach thirty-five million euros or 7% of worldwide turnover.

Digital Omnibus on AI, adopted June 2026; EU AI Act, Articles 50 and 99

The relief is real. It is relief on the hardest obligations, not on the ones an independent house is most likely to breach by accident.

What a house of this size needs: a charter for how the tools may be used, a register of what the house refuses to let them do, and a record of who has been trained. Three documents. A morning’s work with someone who has done it before.

Ask instead

What in my house is being done by a good person at fifteen per cent of their skill and eighty per cent of their day?

Find three of those. Free up the hours. Give them back to your best people and your best clients. The rest of the work in this paper is sequencing.

A note about who is writing this

Twenty years inside the luxury industry, thirteen of which were at executive level in high jewellery and watchmaking. A decade sparring founders.

I spent those years watching how the most accomplished founders carried what they built and what it cost them. ELIA Atelier exists because independent houses serving wealthy and demanding private clients have a once-in-a-decade window in front of them, and most of them are going to miss it. Not because they lack the budget. Because they lack someone who understands both their world and this one.

The practice reads the house, names what must be protected, and builds the instruments that protect it. The Luxury AI Report is part of that work rather than a piece of marketing attached to it. I would rather publish what the engines say and be wrong in public than sell a diagnosis nobody can check.

ELIA Atelier is the AI visibility practice for the luxury industry, founded by Alexandre Olive and based in Genève, Switzerland. Twenty years inside the luxury industry, thirteen at executive level in high jewellery and watchmaking.

Alexandre Olive