The answer is not better SEO and it is not a prettier website. It is packaging what you know so it answers a question inside the tool your customer already has open, and here is exactly how that works now.
The Thing Nobody Announced
Tuesday at OpenAI DevDay, more than twenty things shipped. Cheaper models. Faster models. Autonomous agents with their own cloud computers.
The item that actually matters to your business got about ninety seconds of stage time.
Plugin extensions. Developers can now build native experiences that live inside ChatGPT, in front of 1.2 billion weekly users. Sign in with ChatGPT went live across sixteen partners. An OpenAI Marketplace opened that lets enterprises spend money they already committed to OpenAI with thirty-two outside software vendors, Figma and Adobe and Salesforce and Harvey among them.
So here is the direct answer to the question in the title. You get in front of customers who never leave ChatGPT by putting your knowledge where the question gets asked instead of where you wish it got asked. ChatGPT stopped being a tool people use and became shelf space people can build on. The winning move is no longer "drive traffic to my site." It is "be the thing that answers the question, inside the app they already trust."
Candidly, this reframe took me a while to accept, because I spent twenty years in an industry whose entire job was getting people to come to your website.
That job is not harder now. That job got replaced.
Key Takeaways
- Distribution for a small business shifted from "get people to my website" to "stand next to somebody else's customers," and the AI platforms just opened the shelf.
- In the first four months of 2026, 68.01 percent of US Google searches ended without a click, according to SparkToro and Similarweb, up from 60.45 percent in 2024.
- Platform ecosystems reliably become larger than the platform: IDC projects Salesforce partners will earn $6.19 for every $1 Salesforce earns by the end of 2026.
- Shelf space is not the same as getting picked, and the history of app stores proves it: 91 percent of App Store apps had no organic discovery at all by 2016.
- The same logic applies to expertise, not only software. If your knowledge only exists on your website, you are still knocking on doors.
The Problem: You Were Never Going to Win the Traffic Game
Let me name the thing honestly.
For three years, the distribution problem for a small product or a small practice was brutal and completely unchanged. You had to go get every single user yourself. One post, one ad, one referral, one cold email at a time.
I know this from the inside. I built a digital marketing agency to seven figures and then exited it in 2023. The entire business model was solving this problem for other people, and I will tell you what nobody in that industry says out loud: it was getting less winnable every year, and we all knew it.
Then it got worse in a way nobody planned for.
Search stopped sending traffic. Not slowed. Stopped, in a measurable, one-directional way. And the reflex response from most business owners I talk to is to work harder at the thing that is no longer working. More blog posts. More keywords. A site redesign. A new funnel.
Here is the thing. Effort applied to a closed door is still just effort.
I watch entrepreneurs in our 500,000 member community do this every week. Smart people, real expertise, genuinely useful knowledge, pouring it into a website that fewer and fewer humans will ever load. They are not lazy and they are not behind. They are following instructions that expired.
The instructions expired because the customer's starting point moved. Your customer does not open a browser and type a question anymore. They open an assistant and ask it. And that assistant either knows about you or it does not.
That is the whole problem, stated plainly. The question changed from "where do I host this" to "whose customers am I standing next to."
The Evidence: The Front Door Closed and the Side Door Opened
I do not want you taking my word for any of this. Look at the numbers.
The open web is no longer getting clicks. Rand Fishkin at SparkToro, working with Similarweb's clickstream panel, found that 68.01 percent of US Google searches in January through April 2026 ended without a click. In 2024 it was 60.45 percent. Ten years ago it was around 45 percent. That is the fastest acceleration of that trend in a decade.
And the AI tools are not sending you replacement traffic. This is the part people get wrong when they hear "AI referrals." In the same report, Fishkin notes that AI tools send less than 1 percent of all traffic out. Ahrefs tracks traffic to more than 75,000 opted-in domains, and Google's share of traffic to those sites fell roughly 22 percent between June 2025 and May 2026.
Read those two together and the conclusion is uncomfortable but clear. Nobody is coming to your website. Not from Google, and not from ChatGPT either. Being cited is the win. The click is not coming back.
Platform ecosystems get bigger than the platform. This is the good news and it is well documented. IDC's study for Salesforce projects the partner ecosystem will earn $6.19 for every $1 Salesforce itself earns by the end of 2026, creating 9.3 million jobs and $1.6 trillion in new business revenue. In 2019 that same ratio was $4.29. The people standing next to Salesforce's customers made more money than Salesforce.
Shopify tells the same story from a smaller angle. App developers on Shopify have earned more than $1 billion, and since January 2025 Shopify lets developers keep 100 percent of their first $1 million in gross app revenue. Platforms pay to fill their shelves.
The new shelf is being built right now, in public. Model Context Protocol, the open standard for connecting outside knowledge and tools to AI assistants, went from roughly 100,000 monthly SDK downloads at launch to about 97 million monthly downloads by March 2026. There are more than 10,000 active public MCP servers. OpenAI, Google, Microsoft, AWS, and Salesforce all shipped support inside thirteen months.
Look at what launched on Product Hunt today for a concrete picture. CrawlRaven MCP joins Search Console, GA4, keyword data and a site crawl into one impact-ranked plan. It has no app. You query it from inside Claude, ChatGPT, or Cursor. Nobody is being asked to visit a website.
Now the warning, because I am not going to sell you a clean story. Shelf space is not the same as getting picked. Adjust's research found that by 2016, 91 percent of App Store apps were "zombie apps" with no organic discovery, up from 70 percent in 2013. And of the more than 3 million custom GPTs people have created, roughly 159,000 are public and active. Most things put on a shelf are never touched.
That is not a reason to skip the shelf. It is a reason to be specific about what you put on it.
What Actually Works: Package the Judgment, Not the Content
Here is what changed for me, and it started with an accident.
I never set out to build on somebody else's platform. I built a free Facebook community because it was the cheapest place to talk to entrepreneurs. It grew past 500,000 members. And somewhere in there I realized I had accidentally done the correct thing: I put my expertise where my customers already were, instead of asking them to come find it.
Every piece of my business that works today runs on that logic. The community sits on Facebook. The training sits where people already watch video. None of it required somebody to remember my domain name.
So when the AI platforms opened their shelves this year, the move was obvious to me in a way it would not have been three years ago.
The mistake most people make is assuming this only applies to software. It does not. It applies to expertise, and expertise is what most of my audience actually sells.
Think about what you know that a general model does not. Not facts. Facts are commodity. I mean the judgment calls. The thing you say to a client in month three that saves them from the mistake everyone makes in month four. The reason you price a job the way you price it. The five questions you ask before you will quote anything.
A general-purpose model gives an average answer to those questions, because it was trained on average answers. Yours is not average. That gap is your shelf space.
The practical form this takes right now is a skill file, a custom GPT, or an MCP server. Do not let the words scare you. A skill is a text file with your instructions and your knowledge in it. That is genuinely all it is. I have written more than a hundred of them and the hard part has never been the technology. The hard part is deciding what you actually know that is worth encoding.
The test I use is simple. Can somebody get a better answer to a real question because your knowledge is in the room? If yes, package it. If it is just your marketing rewritten, leave it off the shelf, because it will be one of the 91 percent nobody touches.
One more thing, and this is the part that matters most for a small business. You do not need the whole shelf. You need one question, asked often, in a category where a generic answer costs somebody real money. That is a defensible position and it is available to a one-person business today in a way it was not available in the App Store gold rush of 2009.
Seven Steps to Get On the Shelf
1. Write down the five questions you get asked before somebody buys. Not the questions you wish they asked. The real ones, in their words, from your inbox and your sales calls. These are the queries your customer is now typing into an assistant instead of into Google.
2. Ask a general AI those five questions and read the answers critically. Where is the answer generic, dated, or flatly wrong for your market? Write down the gap. That gap is the only asset that matters in this whole exercise.
3. Run this prompt to find your packageable knowledge. Use it with whichever assistant you already pay for.
[The Job]
Find the five questions my customers ask an AI assistant that my expertise could answer better than a general-purpose model.
This is for: [YOUR BUSINESS TYPE] serving [YOUR CUSTOMER].
It matters because: if I know those five questions, I know exactly what to package so it shows up inside the tool they already have open.
[The Background]
Here is what you need to know: I sell [YOUR PRODUCT OR SERVICE]. My customers are [DESCRIBE THEM IN ONE SENTENCE]. The three problems they hire me for are [LIST THREE]. The judgment calls I make that most people in my field get wrong are [LIST TWO]. I am attaching [YOUR FAQ PAGE, PAST PROPOSALS, OR SUPPORT EMAILS].
Do not use: generic marketing questions, keyword research, or anything a competitor could copy from a blog post.
[The Deliverable]
Return: a table with five rows.
Must include: the question in the customer's own words, why a general AI answers it poorly, the specific thing I know that would make the answer better, and the smallest packageable form of that knowledge.
Optimize for: accuracy.
Written from the perspective of a [SKEPTICAL BUYER] talking to their own AI assistant.
[The Questions]
Ask me any questions you have.
4. Build the smallest possible version first. One skill file or one custom GPT that answers one question well. Not a suite. A single useful thing beats a broad thing nobody finishes reading.
5. Put it where your customer already is, not where you want them. If your buyers live in ChatGPT, build for ChatGPT. If they live in Claude or in a spreadsheet, build there. The platform choice is a research question about your buyer, not a preference question about you.
6. Keep a front door that is yours. Your email list, your community, your phone. The platforms will change the rules, because they always do. Standing next to somebody else's customers is a distribution strategy. It is not a substitute for owning the relationship.
7. Give the first one away. Free distribution of something genuinely useful is how you get invited back. That is the same rule that built a 500,000 member community, and nothing about AI changed it.
Frequently Asked Questions
Do I need to be technical to build a skill or a custom GPT?
No. A skill file is a text document containing your instructions and your knowledge. You write it in plain English and the AI reads it. If you can write a training manual for a new hire, you can write a skill. The technical part is copying a file into a folder.
Is this just SEO with extra steps?
No, and the difference matters. SEO tries to earn a click to your site. This puts your answer inside the assistant, where the answer is consumed and no click happens. SparkToro's data shows AI tools send under 1 percent of traffic out, so optimizing for the click is optimizing for a shrinking prize.
Should I abandon my website?
Absolutely not. Your site still feeds what AI assistants say about you, and it is the one property where you set the rules. Rand Fishkin makes this exact point: even when traffic falls, your site shapes AI answers. Keep it. Just stop treating traffic to it as the scoreboard.
Which platform should I build on first?
Whichever one your specific customers already open daily. ChatGPT has the largest reach at 1.2 billion weekly users, so it is the default answer for consumer-facing businesses. If you sell to developers or technical teams, Claude and Cursor may reach your buyer more directly than raw user counts suggest.
Is it already too late, like the App Store in 2012?
Not yet, though the window behaves the same way. Adjust found 91 percent of App Store apps had zero organic discovery by 2016. The advantage today goes to narrow, specific expertise rather than broad tools, which is exactly the kind of thing a small business has and a large one does not.
The Question Changed, So Change Your Answer
I opened with the thing nobody announced on the DevDay stage. Let me close with why it kept me up that night.
For three years I have watched good people lose a fight they never had a chance to win. They built the site. They wrote the posts. They did what the last decade of marketing advice told them to do, and the door closed anyway. That is not a failure of effort and it is not a character flaw. It is a change in the terrain, and nobody sent a memo.
Here is the memo.
Nobody is asking you to come to their website anymore, and that means nobody is coming to yours. The distribution problem you have been fighting did not get easier. It got replaced with a different problem, and the different problem is one you can actually solve, because it rewards the thing you already have: specific, hard-won judgment in a narrow lane where a generic answer costs somebody money.
You do not need a development team. You do not need permission. You need to decide what you know that is worth encoding, write it down, and put it where the question gets asked.
If you want to work through that with people doing the same thing, that is what AI Insiders is for, at $247 a month. We build these in public, we share what breaks, and nobody there is pretending this is simple.
But whether you join us or not, do the first step this week. Write down the five questions. Ask an AI to answer them. Read what it gets wrong.
If the thing you know can answer a question inside a tool your customer already has open, you are on the shelf. If it only exists on your website, you are still knocking on doors.
P.S. I built a seven figure agency whose whole job was getting people to visit websites, and I am now telling you to stop optimizing for that. I would rather be useful than consistent.
About the author
Jonathan Mast is the founder of White Beard Strategies, an AI education and coaching company for entrepreneurs, and the creator of the Perfect Prompt Framework 2.0. He serves a 500,000 member community of entrepreneurs learning to use AI without a technical background, speaks regularly on practical AI adoption, and has written more than a hundred AI skill files. He founded White Beard Strategies in 2023 after exiting the digital marketing agency he built to seven figures, which is a large part of why he is unsentimental about website traffic. He lives in Baldwin County, Alabama.
Sources
- In 2026, Less than One Third of Google Searches Still Send a Click, SparkToro
- ChatGPT vs Google traffic tracker, Ahrefs
- 9.3M Jobs and $1.6T in Revenue: The Salesforce Business Partner Ecosystem
- New Study Finds Salesforce Economy Will Create 9.3 Million Jobs and $1.6 Trillion in New Business Revenues by 2026, IDC via PR Newswire
- The developers behind Shopify's $1.3 billion app ecosystem, Shopify
- Revenue share for Shopify App Store developers, Shopify Developer Docs
- MCP Hits 97M Downloads: Model Context Protocol Guide, DigitalApplied
- Model Context Protocol, Wikipedia
- Lacking organic discovery, 91% of games and apps on App Store are zombies, PocketGamer.biz citing Adjust
- GPT Store & Custom GPTs: Complete Business Guide 2026, DigitalApplied
- Introducing the GPT Store, OpenAI
