Your next user is describing your product to a chatbot.
They don't know your name. They're typing the problem - the workflow that keeps breaking, the spreadsheet nobody wants to own, the thing their team does by hand every Friday - and something is going to come back as the answer. Getting your product into that answer is a different job from ranking for a keyword, and it's the one this playbook is built around.
Book Your Impact WeekThe SEO and AI visibility playbook
GTM Engineer allocation
50-100%
The flattest load on your plan. Half a week holds the publishing slot and the rewrite queue open; the top of the range is for the months with original research in them.
- Keywords chosen on what a signup is worth, not on what looks easy to rank for.
- Pages built to be found by search engines and quoted by AI assistants.
- A monthly read on what the assistants say when somebody describes your problem.
Where this playbook comes from.
Software gets found differently now. A person with a problem describes it to a search engine or an assistant, reads whatever comes back, and signs up for something before they've spoken to anybody. Your product either appears in that moment or it doesn't exist yet.
This is the slowest experiment on a Designli go-to-market plan and the only one still bringing you users a year after the engagement ends. It earns its slot when people are already searching for what your product does, and when you have enough runway to wait a quarter for the first honest read.
When your category is so new that nobody has a word for the problem yet, there's no question for you to be the answer to. Your engineer will put a channel that creates demand ahead of one that captures it, and say so during the Impact Week rather than in month three.
You can't train ChatGPT.
Anyone telling you otherwise is selling something that doesn't exist. There's no submission form, no opt-in, and no way to edit what a model already believes about your product. What you can change is what it finds when it goes looking, because every assistant worth caring about reads the live web while it's answering.
So the work is making sure the answer to your users' question sits on a page an assistant can reach, written so it can be lifted cleanly and attributed to your product by name, and repeated on the third-party sites it already trusts. Change the source material, ask again, and watch the answer move. That loop is the whole mechanism, and it's measurable.
It's also the honest reason this channel takes a quarter to read. The work and the payoff sit three months apart, and then it keeps paying after you stop. Every other channel on your plan stops the day you stop feeding it.
How an SEO experiment actually runs.
The loop is the same one every channel on your plan runs. What's different is the clock - the hypothesis you write in week one gets its answer in month three.
1. Start from what your users type, not what you call it.
People search the problem in their own words, which are almost never your product category. Your engineer builds the list from how they actually describe it, then keeps only the terms where somebody typing them is close enough to signing up to be worth chasing.
2. Make the answer extractable.
The answer goes in a single paragraph before the first heading, with your product named in the sentence rather than implied, so an assistant can lift it and credit you. Every page after that is built the same way - the question a user would actually ask, then the answer, in that order.
3. Start publishing before there's anything to buy.
One piece a week, on the same day every week, starting the week the research is done - even if the product is still a waitlist. Nothing compounds until something is published, so nothing waits for launch.
4. Tell the crawlers they're welcome, and hand them the structured data.
Your robots file names the AI crawlers explicitly instead of trusting a wildcard, and the server logs get read to confirm the ones you meant to allow are actually turning up - the usual culprit isn't a policy decision, it's a firewall rule nobody knew about. Product, organization and article markup goes in so machines can tell what your software does.
5. Rewrite what already works before writing anything new.
A page that's been bringing you signups for two years already has the authority a new page spends a year building, so the highest-return work is usually a rewrite: same URL, current argument, retitled to the question people type now. When two of your pages compete for one query, one gets absorbed into the other.
6. Re-run the same questions and watch the answer move.
Every month the same user questions go to ChatGPT, Gemini, Perplexity and Google's AI Mode, several times each, recording how often your product comes up rather than where it ranks - ask twice and the list changes, so anyone selling you a rank inside an assistant is selling you noise. What matters is the description drifting toward what your site now says, and citations arriving from more than one assistant.
A typical ramp-up plan.
This is the only experiment on your plan still working a year after it's paid for, and the only one that asks you to wait a quarter for the first honest read. Your engineer spends that quarter buying assets you own outright - the research, the crawl surface, and a library of pages that keeps bringing users in after the engagement ends. The clock below starts the day the first page goes up, not the day the plan is approved.
The first four weeks are all foundation.
The keyword research gets done in your users' language, the crawl surface goes in behind it - sitemap, search console, structured data, an AI-crawler policy - and the first pages publish before the month is out. You end the month owning a research file, a technically clean site, and a publishing slot that has already run.
By week twelve you can see which pages have legs.
Ten to twelve pieces are live and the oldest have been crawled at least once. The long tail starts to widen as pages nobody predicted start bringing people in, and your engineer stops guessing which problems to write about and starts writing more of what already proved out.
From month four the archive does the work.
The rewrite queue takes over from net-new, because a page that already has authority moves faster than a page with none. Assistants start naming your product from more than one engine, and the cost of the next signup falls instead of climbing.
What you can measure, and what nobody can.
Every week the same slot runs, scoped to the same seven days so nobody can quietly switch to a trailing thirty. New users first, then organic sessions, then how the content is feeding signups - each stated as a direction rather than a percentage, because on a weekly scale a percentage is theatre.
Then the part that didn't exist three years ago. Which of your pages an assistant cited, read out by name. How many citations, split by engine. And the monthly prompt check: the questions a real user would ask, re-asked across the assistants, recording whether your product came up and the exact words used to describe it. That last one is the closest thing there is to hearing your market explain your product back to you.
Competitive position gets reported once a month and never as a lone number. Your score goes on the screen next to three named competitors, because an absolute score means nothing on an index that moves under everybody. One search-engine update last summer knocked a whole category down at once, and the only honest read was who fell least.
Two things nobody can measure, and we'd rather say so. A user who gets their answer inside an assistant and never clicks is a real result you'll never see in analytics - the only proxy is asking the assistant yourself. And you don't get to choose who arrives or how many. Outbound lets you decide that six hundred named companies hear from you this month; search hands you a session, not a name, and you take whoever the answer sends.
That's the trade. In exchange you get the one channel that keeps producing after the spend stops - a page published in month two is still bringing you signups, and still being quoted to people describing your problem to a machine, in year two.
Let's find out how your users describe the problem you solve.
It starts with a free 30-minute call. Tell us what your product does and who it's for, and we'll tell you whether anyone is searching for it in those words, what it would take to own those answers, and which channel we'd run ahead of this one.