Vision PaperVP-2026-001
IssuedSeptember 2026
Midnight × GYOTAKU Protocol

The trust layer for a ¥7,000 trillion AI economy.

By 2040, AI will account for about 20% of world GDP — ¥7,000 trillion (Masayoshi Son, SoftBank World 2026)
Then
Did the person who wrote that post actually buy it?
AI Advertising Midnight Agent Commerce GYOTAKU Protocol Pranburi, Thailand
Vision Paper · Not a technical report
2025
73.2%
Internet share of ad spend, 2025
2040
100trillion
AI agents by 2040 (Masayoshi Son)
Upside
up to¥350T
AI ad market by 2040 (GYOTAK estimate)
Status
Mainnet
Purchase records live on Midnight
Abstract / 要旨

Advertising budgets are being paid to people who never bought the product. Now that AI can invent the person as well as the post, a reader cannot tell the real from the fabricated by reading. GYOTAK pays the advertising budget to the buyer, lets only buyers refer, and writes the backing records to the Midnight blockchain: where the fish was caught, whether the cold chain held, that someone really bought it, and that the referral link in the post belongs to that buyer. Anyone — or any AI agent — can check all four without trusting GYOTAK. This paper sets out the market that shift creates, the mechanism, and what it does not prove.

01 / Who Wrote That Post

Who wrote that post?

Paying for attention, not for truth.
Ad budgets pay whoever got attention — not whoever actually bought.

Wimbledon, 2025. Tennis fans stopped at one woman’s posts. Mia Zelu. The writing carried the feel of the grounds; the photographs looked real. She does not exist. An AI made her.

The Spanish AI model Aitana López has been reported to earn up to $11,000 a month from brand deals.

They never buy the shoes or the clothes. What they wear is generated, swapped on and off. There is no guarantee they produce the next buyer. And still, advertising money is paid.

For now, this kind of advertising still works. But by its nature, it cannot last.

AI has pushed the cost of making content close to zero. Anyone can produce any number of convincing posts. The number of posts explodes, but the amount a person can look at in a day does not. What rises instead is the cost of being found.

As the cost of being found keeps rising, so does what manufacturers must pay. Yet the number of real buyers those ads produce does not rise. AI influencers don’t buy, and many of their followers only look. Brands pay more and get no more customers. Sooner or later, manufacturers lose any reason to pay for this.

Meanwhile, the shopper itself is shifting from people to AI agents. An agent will gather information on its user’s behalf, check it, select the best, and deliver only what fits. What gets chosen is not the best-looking ad, but the record that can be verified. Advertising paid by the view loses its viewers.

And now, AI can write the same message a thousand different ways in a second. Telling the real from the made-up just by reading is no longer possible.

In a world where making things costs nothing, what keeps its value is what cannot be made — the fact that someone really bought.
02 / Advertising Changes Hands

Advertising changes hands every twenty years

The next twenty years, in fourteen.
The internet took advertising from 4% to 73% in twenty years. Next, AI will change today’s internet advertising.

In 2005, internet advertising was roughly 4–6% of global ad spend. It passed television in 2017 and reached 73.2% in 2025. In twenty years, the medium at the centre changed.

Year Internet share of ad spend Source
2005about 4–6%ZenithOptimedia
201736.9% (overtook TV)Zenith
202573.2% (81.6% incl. digital extensions)WPP Media

The global ad market was about $1.08–1.14 trillion in 2025, of which internet advertising was about $790 billion (≈ ¥119 trillion).

On 14 July 2026, SoftBank Group’s Masayoshi Son forecast that by 2040 world GDP will reach ¥37,000 trillion, that AI will account for about 20% of it — ¥7,000 trillion — and that 100 trillion AI agents will be running. “It will be an agent-centred society, not a human-centred one.”

In 2030, AI advertising will stand roughly where internet advertising stood in 2005: WPP Media projects generative-search advertising to grow from $5.1 billion in 2026 to over $100 billion by 2030, about 6% of the 2030 market. Using the internet’s own path as the yardstick gives the following for 2040.

Scenario Reasoning AI share AI ad market in 2040
Conservative Same pace as the internet about 33% ≈ $1.0 trillion (≈ ¥150 trillion)
Base Faster than the internet (no new devices or networks needed; it runs on top of the existing internet) about 50% ≈ $1.5 trillion (≈ ¥230 trillion)
Upside AI takes over the internet’s share of advertising about 73–82% ≈ $2.2–2.5 trillion (≈ ¥335–375 trillion)
Note: The 2040 global ad market (≈ $3.06 trillion) extends WPP Media’s 2026 figure of $1.3 trillion at its five-year CAGR of 6.3%. ¥150 = $1. This is a GYOTAK estimate built from WPP Media and Zenith data, not a published forecast of 2040 AI advertising.
Up to ¥350 trillion — and that counts only the advertising shown to people.

The purchasing that 100 trillion agents decide, after checking, is many times larger. An agent does not choose on atmosphere. It chooses on records it can verify.

03 / The Age of Earning by Buying

The age of earning by buying

Buy. Share. Earn. Repeat.
Buying is no longer just spending money.
The buyer becomes the earner — and every referral becomes an asset that grows.

Until now, advertising money went to agencies and influencers. The buyer paid, enjoyed the product, and that was the end of it.

GYOTAK turns this around. The buyer shares their own experience, and when a new customer buys because of it, the buyer earns a reward.

And the reward does not stop at one. For as long as that customer keeps buying, every repeat order keeps paying the person who referred them. One referral grows into an ongoing income. A referral is not a one-off ad — it is an asset you grow.

So buyers don’t just eat, enjoy and move on. They start to find satisfaction in finding the next customer, and want to tell people about a great meal. And the person they referred can move to the earning side in exactly the same way.

Consumers recommend to consumers. Not because a brand paid them to, but because they bought it and liked it. Because the person they refer gets the very same way to earn once they buy, so it is easy to recommend. And because their own income keeps coming. That motive creates a flow of advertising that has never existed before.

Only people who actually bought can refer. No follower count, no application.

A motive backed by the record

Does this person really keep buying good things? Each purchase they choose to speak about is stacked on the blockchain, recording what they bought and how many times. Behind the recommendation is the recommender’s own buying history.

And anyone can check it directly on the blockchain — without taking GYOTAK’s word for it, or the recommender’s. No one can rewrite the record afterwards. A recommendation that doesn’t ask to be believed. That makes it credible on a completely different level.

04 / Just Ask Your AI Chat

Just ask your AI chat, and it’s ordered

No login. No crypto. Just ask.
The blockchain carries the trust. Nobody using it has to know that.

However good a trust system is, it won’t spread if it makes people do extra work. That is the one thing GYOTAK cares about most.

Buyers don’t need to know anything about blockchains. They don’t need to hold any cryptocurrency. They pay with ordinary money (in Thailand, a PromptPay bank transfer), and referral rewards are paid the same way. The experience feels exactly like shopping always has.

What changes is how easy it is. Shopping with GYOTAK happens entirely inside the AI chat you already use every day (such as Claude). No account on GYOTAK’s site, no ID, no password. Just say “Two packs of sashimi-grade tuna, please,” and the fish is chosen and the order and payment instructions follow.

Online shopping used to mean opening a site, signing up, logging in, filling a cart and typing your address before you could buy. With GYOTAK, you just talk. Shopping that is easier than online shopping starts here.

The blockchain works quietly in the background. All the buyer sees is their usual chat, the fish that arrives, and a “verify” button.

No extra work for the person posting

The post is written by the AI the buyer already uses. GYOTAK returns only the facts — which fish, which date — and what may and may not be claimed. It never ghostwrites. So the post comes out in that person’s own words and their own impressions, and they can edit it if they want. It is easier than posting normally.

Through all of this, the person posting never thinks about the blockchain. They never need to know it is there.

And for those who want to go one step further: the photo

For people who want more, there is the capture feature. Photograph the fish through GYOTAK’s capture page, opened from the chat, and the photo’s own fingerprint (a hash), the time it was taken and the broad area it was taken in are written to the blockchain. The exact location stays hidden by zero-knowledge proof; only “when, and roughly where” remains. The page only accepts a photo taken then and there with the camera, so an image prepared in advance cannot be used.

This was built on a simple expectation: AI agents will come to value photos a human actually took. In a world flooded with AI-generated images, a photo that can prove it was taken carries value on its own.

Using it is entirely optional — posts work without it. Either way, there is no extra work.

Invisible by design

The most advanced trust layer, delivered so no one has to notice it. Technology behind, convenience in front.

05 / The Referral Loop

The referral loop

The right to refer is issued by the purchase itself.
The right to refer is issued by the purchase itself.

Figure 1 — the referral loop

Step 01Order

You order in an AI chat window. No ID, no password

Running
Step 02Payment

On confirmation, a referral link of your own arrives automatically

Running
Step 03Record

The fact of the purchase is written to the blockchain with your name left out

Running
Step 04Post

Your own AI writes the post. GYOTAK returns only the facts and what must not be claimed. We never ghostwrite it

Running
Step 05Claim

If you choose to, the account you post from and your own referral link are stamped into the same record

Running
Step 06Next order

A reader orders through that link, and the loop closes

Running

A referral link, once received, can be used forever. You can keep posting it on social media.

But the blockchain also records when you last bought. If someone who bought once, years ago, keeps posting the same experience, any reader can check it — and see that they are only talking about something from long ago.

People who keep buying are different. Every purchase adds a new record, and the fact that they are still eating it, and buying it again and again, stays visible to everyone.

Buying again is what makes the voice credible. A referrer is trusted not for how well they advertise, but for the fact that they keep buying. And readers can check this on the blockchain — without relying on GYOTAK or on the referrer.

Referrals to yourself are automatically excluded from rewards.

A one-time experience from long ago is exposed by the record. Only the voices of people who keep buying build trust over time.
06 / Four Records Behind One Post

Four records behind one post

Hide what must stay hidden. Prove what must be proven.
Behind “it was delicious” sit four records anyone can check.

Figure 2 — four records behind one post

01 When and where it was caught TR-2026-012
Kept hidden
The exact fishing position
Proven
This fish was landed in this region, on this date
02 Whether the cold chain held TR-2026-011
Kept hidden
The raw temperature readings and the threshold
Proven
The standard was met
03 Whether it was really bought TR-2026-015
Kept hidden
Who the buyer is
Proven
Someone did buy this lot
04 Whether the referral link belongs to that buyer TR-2026-016
Kept hidden
—
Proven
The link in the post and the link in the record are the same

You do not have to ask GYOTAK to check any of this, because the records sit on the blockchain rather than in our database. There are two ways in. You can connect the MCP server GYOTAK publishes and ask an AI to verify a record; or you can query Midnight’s public data yourself and decode it yourself, which involves GYOTAK not at all. Both routes are written out step by step in TR-2026-012. In an end-to-end test, an AI read the region and the species straight off the chain and correctly concluded that the exact position cannot be recovered.

Copy the post, swap in your own link, and it stops matching the record.

That post fails a check any reader can run, because only the buyer could have produced that pairing in the first place.

The difference between an AI that can check and one that cannot is stark. An AI connected to GYOTAK’s records read the region and species correctly from the chain — and correctly concluded that the exact location could not be recovered. A chat AI with no way to reach the records answered the same question by inventing a plausible region and species (TR-2026-012). In an age when agents do the shopping, what matters is not well-written text but records an agent can actually go and read.

Gyotaku. A Japanese technique: press a fish onto paper and take its exact shape. What the paper takes cannot be edited afterwards. The blockchain is the paper; the proof is the fish. More on the name →
07 / A Post You Can Check

A post you can check is worth more

What an agent can verify, an agent can pay for.
A post you can check is worth more than a post you cannot.
A usual influencer post A GYOTAK post
Did they really buy it? No way to tell It is in the record
Who gets paid Someone who never bought it can The buyer
Do they keep buying? No way to tell One more record per purchase
Who wrote it It may have been ghostwritten The buyer (GYOTAK never ghostwrites)
If it is copied Indistinguishable It stops matching the record
Can an AI agent choose on it? Only by reading the wording It can read the record and decide
08 / What This Means for Midnight

What this means for Midnight

Every trusted recommendation runs on DUST.
Ad budgets disappear once spent. NIGHT keeps generating DUST for as long as it is held.

Until now, a brand’s advertising budget was an expense that disappears once spent. Pay per view, then pay the same again next year.

In the GYOTAKU model, every record of trust is written using DUST. And DUST keeps being generated every day, for as long as NIGHT is held. For a brand, NIGHT becomes not an expense that disappears, but an asset that keeps producing the power to write records. That gives brands a reason to move their ad budget from a yearly expense into NIGHT they keep.

Suppose half of the 2040 AI advertising market moved to this model, with brands holding NIGHT equal to one year of their ad budget.

Sector Assumption Demand to hold NIGHT
AI advertising 50% of the 2040 AI ad market (up to ≈ ¥350T) moves over; brands hold NIGHT equal to one year of ad budget ≈ $1.2T

That is roughly 2,900 times NIGHT’s current market cap (≈ $400M).

And that is advertising alone. Advertising is far from the only reason to need NIGHT.

The market for tokenized financial assets is projected to reach $18.9 trillion by 2033 (Ripple and BCG). Bonds, funds, deposits, trade finance. Much of this activity cannot be made public — counterparties, amounts and client data must stay confidential — and yet it must be provably correct. Proving while keeping things hidden is exactly what Midnight does best.

Finance, healthcare, government, supply chains — there are countless fields beyond advertising that must prove what is correct while keeping secrets. If all of them write their records with DUST, demand to hold NIGHT will reach far beyond the advertising figure alone.

Look at it from another angle: work backwards from how people actually behave.

Suppose half the world’s population — about 4.1 billion people — shopped this way. Each person spends money twice a day, and each purchase writes two records (the purchase itself, and its link to a referral). One NIGHT generates up to 5 DUST per week.

Assume each record costs 40–50 DUST. On mainnet, each of GYOTAK’s catch records costs about 40 DUST.

DUST per record NIGHT required vs. total supply (24 billion)
40 DUST ≈ 920 billion ≈ 38×
50 DUST ≈ 1.15 trillion ≈ 48×

The NIGHT required comes to about 38–48 times the total supply. Put the other way round: even the entire supply of NIGHT could support only about 1% of humanity. The shortfall itself shows how much demand this model creates for Midnight.

Scarcity means selection.

When the capacity to write records is limited, companies adopting this model will weigh the cost of securing NIGHT against the value the records create. The transactions where a single record creates the most value will adopt it first: high-value products, high-value deals, and high-value experiences such as travel. Trust will gather on Midnight in order, starting with the transactions where proof is worth the most.

That is why moving early is an advantage. Companies that hold NIGHT early and secure their capacity to write records will keep writing records of trust steadily, even as demand grows. Those who arrive later will be competing for what capacity remains.

Records of trust will choose Midnight in order of value. Those who prepare first will hold that place.

Why would it move? Because today’s advertising and today’s AI share the same flaw. Advertising pays for views with no way to check whether the voice is real. AI can produce articles, photos, videos and entire media outlets that look exactly like the real thing.

Technology to prove what is real is already arriving. In September 2026, Apple announced “Apple Reference Image,” which proves that a photo taken on iPhone 18 Pro or Pro Max comes from the actual camera sensor, not from AI. The sensor signs the pixels the moment they are captured, and Apple’s cloud verifies them. The camera and software industries also have C2PA, a standard for digital signatures. These are strong designs, but what they prove stops at “taken by a real camera, within this window of time.” They do not prove where it was taken, what the scene shows, or whether the person who took it actually bought what is in it. And their trust starts with the company that signs. Verification, and revoking a proof, are in that company’s hands.

Midnight adds the missing layer. Write a photo’s fingerprint to Midnight, and the fact that “this photo existed at this time” becomes a public record no company can erase. With zero-knowledge proofs, the area where it was taken can be shown while the exact location stays hidden. Link it to a GYOTAK purchase record, and it becomes “a photo really taken by someone who really bought.” Apple and C2PA show that it was taken by a real camera; Midnight shows it has not been erased and belongs to someone who really bought. Together, they deliver trust that doesn’t require trusting any company.

Systems that record whether a photo is real, or whether a reviewer really bought, already exist around the world. What sets GYOTAK apart is that it connects them. It starts from records of the goods themselves — the catch, the freezer temperature — and runs through the purchase, the buyer’s public voice, and the referral reward as one continuous record. And it proves all of it while keeping hidden what must stay hidden: names, fishing grounds, where a photo was taken. Not the parts, but this continuous whole, is the GYOTAKU Protocol.

Hidden: who the buyer is, exactly where the boat fished, the temperature readings themselves.
Proven: it came from this region; the temperature standard was met; someone really bought it; this referral link belongs to that buyer.

If you can only do one of the two, the design collapses. Publish everything and the buyer’s privacy, the fishing grounds and the method are all gone. Hide everything and nobody can check anything.

What makes this possible is Midnight’s design itself. Zero-knowledge proofs hide what must be hidden and prove only what must be proven. Smart contracts set the rules for each record. And DUST, which cannot be traded or speculated on, keeps the cost of recording predictable. To our knowledge, Midnight is the only chain designed with all three together from the start. That is why, today, Midnight alone can aim for the position of trust layer for AI advertising.

Today, only Midnight can build this while keeping hidden what must stay hidden.

The first companies to realize this will move. They can gather the most trusted customers of next-generation advertising before anyone else. Once one moves, its competitors have to follow. Competition begins, and the number of companies that need Midnight keeps growing.

In crypto, good technology is eventually adopted elsewhere. Even so, rebuilding zero-knowledge proofs, smart contracts and a non-tradable fee resource as one integrated design will take other chains years.

Those years are Midnight’s window. In that time, Midnight can become the first choice as the foundation of trust for the economy outside crypto — advertising and everyday shopping — and build a track record. Records cannot be back-dated. A track record built first is one no latecomer can catch.

The further GYOTAK spreads, the more records live on Midnight, and the more reasons there are to hold NIGHT. GYOTAK’s growth and Midnight’s growth point in the same direction.

Growing Together

Every record GYOTAK writes is one more reason to use Midnight.

These figures illustrate the scale of holding demand. They are not a forecast of NIGHT’s price or market cap, and are not investment advice. World population ≈ 8.2 billion. DUST generation per Midnight’s published design (a cap of 5 DUST per NIGHT, reached in about a week). DUST per record is an assumption based on GYOTAK’s mainnet catch records (about 40 DUST each) and may change with network conditions. Tokenized asset forecast: Ripple and BCG (April 2025). NIGHT market cap: public data from CoinGecko, CoinMarketCap and others (September 2026). ¥150 = $1. DUST is generated up to a cap set by the amount held, and decays when unused.
09 / Running Today, And Not Yet

Running today, and not yet

Status as each technical report states it.
We do not write down as running what is not running.
What is recorded Status Report
The catch (date, species, region; position hidden) Deployed on mainnet; verified on the test network TR-2026-012
Cold-storage temperatures Mainnet TR-2026-006, 011
The purchase (buyer hidden) Mainnet TR-2026-015
Tying the referral link to the purchase Mainnet TR-2026-016
x402 settlement (binding an AI agent’s order to its payment) Live payments on Cardano mainnet; full agent round-trip tested on the test network and ready to deploy TR-2026-003, 004, 005, 008, 014
Telling each box apart by its surface pattern Client-side implementation; test network TR-2026-009, 010
Proving a formula ratio Test network TR-2026-007
10 / From Fish to Every Industry

From fish to every industry

The pattern, not the product.
Industry leaders are starting to say the next era is about trust. GYOTAK is already running the system that proves it.

In 2026, Filip Filipov, CEO of global aviation data leader OAG, told an international travel industry conference that before the internet, people bought trust, not flights. The industry, he said, had taught the world to buy the cheapest option — and should now teach it to buy what can be trusted.

In the age of AI agents, he argued, the winners will not be those who own the customer touchpoint, but those who own trusted information. As its foundation, he named trusted identity, an end to fraud, and information that businesses can share with one another.

This points in exactly the same direction as what GYOTAK has been doing with fish.

What industry leaders say is needed, GYOTAK is already running on mainnet.

Travel: the first expansion

Travel is one of the fields where this model works best.

Connect a travel operator’s booking system to GYOTAK. The traveller books in the AI chat they already use. Payment goes directly to the travel operator; GYOTAK never holds the money. It records only the fact that this booking was paid for.

When the traveller comes home and posts about the trip, behind that post is a record that this person really booked, paid for and took this trip. When a reader books the same trip through the referral link, the travel operator pays a fee to GYOTAK and to the traveller who referred them.

Anyone can write a travel story. AI can write a brilliant one. But someone who never went cannot create the record that story is supposed to sit on. Travel operators win their next customers through the voices of people who really travelled. Travellers earn from their own experiences.

To every industry

Fish or travel, the shape of the system is the same: a record that only the person who had the experience can hold, bound to that person’s public voice, kept where no one can change it.

That shape works in any industry:

Every industry has the same problem: no one can tell a real experience from a manufactured reputation. The GYOTAKU Protocol makes that difference verifiable, without having to trust anyone.

What follows is vision; beyond fish, it is not yet implemented. The travel model is disclosed in TR-2026-016 § 7.
The Pattern

Whatever the industry, the answer is the same: a record only the person who experienced it can create, kept where no one can change it.

魚拓
GYOTAK adapts to its market;
GYOTAKU never compromises on the record.
In a world buried in fraud, people lose their ethics. When it can be verified that the referrer really bought, was satisfied, received the goods and was not deceived, the next person can buy with confidence. Genuine experiences, passed on, keep ethics intact.
VP-2026-001 · September 2026 · GYOTAKU Protocol
ECOSUS CO., LTD. — GYOTAK · Pranburi, Thailand · 0205562030631
gyotak-shop.pages.dev
Appendix / 付録

Sources

Every figure above traces to one of these.
Document IDECOSUS-VP-2026-001
Title (EN)Proof Is the New Impression — The Trust Layer for AI-Era Advertising
Title (JP)証明が、新しいインプレッションになる — AI時代の広告を支える信頼レイヤー
Date2026-09-20
ClassVision Paper — contains market forecasts and vision. For implementation status, the description in each TR governs.
AuthorTakuya Ogura, Chairman, ECOSUS CO., LTD.
LicenseCC BY 4.0
RelatedTR-2026-013, TR-2026-015, TR-2026-016