AI SEO: A Ten-Step Process, and What Actually Changes in Your Work
AI SEO explained as a process, for people who already run SEO. Google states in its official guidance that existing SEO practice remains effective, so this is about adding three things rather than replacing anything. Includes a work-level diff across six job areas and ten steps with completion criteria. In our survey of 400 appliance buyers, 29.5% used AI before purchasing.
Key takeaways
- AI SEO does not replace the SEO you already do. It adds three things to your current work
- Google states in its official guidance that SEO best practices remain effective. This is not a story about throwing anything away
- Demand has moved in Japan too. In our survey, 29.5% of appliance buyers used AI before purchasing, and about 4.7% of traffic to our own site now arrives via AI
AI SEO means adding three things to the SEO you already do
Getting your pages into the candidate set, and quoted accurately, when a generative AI assembles an answer. It adds three things to existing SEO: making pages readable by AI crawlers, writing in a form an answer can use, and measuring how you appear inside AI.
A search results page returns a list of links; an AI answer returns one conclusion. We moved from a world where ranking tenth still left room for a click to one where missing the top three candidates means not existing. That difference determines what has to be added.
What has not changed is the underlying structure: search indexes remain the foundation. A page Google has not indexed cannot become a candidate for Gemini either. Existing SEO therefore becomes a precondition, with the three additions sitting on top of it.
Our overview of LLMO covers the discipline as a whole. This article is written as a process, for readers who already run SEO.
Google states that SEO remains effective
The biggest worry an SEO practitioner has is worth answering first.
If you're already following our image SEO best practices and video SEO documentation, you're already optimizing for generative AI search.
In its guidance for AI features on Search, Google explains that generative AI features are rooted in the core Search ranking and quality systems. AI Overviews, AI Mode and Gemini all stand on the Google index.
Past investment has not been wasted. Existing SEO becomes the precondition, and new work sits on top of it.
What actually changes in the work
Seen at the level of daily tasks, the change is narrower than it sounds. Here it is across six job areas.
| Job area | What you do today | What to add | What does not change |
|---|---|---|---|
| Keyword research | Select keywords by search volume and competition | Write out roughly twenty questions customers would ask in conversation | How you read demand and set priorities |
| Content production | Cover the search intent comprehensively | Make headings a question or a claim, with the answer immediately below. Attach sources and dates to figures | The value of original data, an independent point of view, quality standards |
| Internal linking and information design | Connect related pages through topic clusters | Reconcile the same fact wherever it appears across the site | The cluster structure itself and the design of hub pages |
| Technical SEO | Make pages retrievable and renderable for Googlebot | Confirm key information reads without JavaScript. Review which AI crawlers are permitted | Page speed, mobile handling, crawl efficiency |
| Links and external signals | Manage the volume and quality of acquired links | Include unlinked mentions and placements in comparison and review sites | That third-party evaluation is what carries weight |
| Measurement | Track rankings, impressions, clicks and conversions | Separate AI referrals into their own channel. Record mention rate through prompt measurement | Designing metrics backwards from conversion |
Look at the last column. The fundamentals of keyword research, your content quality bar, internal link design and the technical SEO foundation all continue to work. Only two areas change substantially.
The first is measurement. AI referrals are invisible in a default GA4 setup. Referrers frequently do not carry through, so those sessions land in "direct". Without a configuration that separates ChatGPT and Perplexity into their own channel, you will keep undercounting the results of your work. GA4 alone also cannot tell you how you appear inside AI. Traffic is the outcome; whether you were mentioned at all, and in what position, has to be measured separately.
The second is technical SEO. You used to work against Googlebot, which renders pages before evaluating them. Major AI crawlers such as GPTBot and PerplexityBot do not execute JavaScript and read only the HTML that comes back. On a page where price, stock or specifications are drawn by JavaScript alone, the information effectively does not exist for AI. This happens on pages that rank well in Google. Our article on GEO for e-commerce covers the mechanism with measurements.
Keyword research also changes target. AI does not search the question it was given; it rewrites it into several queries and runs them in parallel. The AI decides the search terms, so they do not match the words the user typed. Switch from winning one keyword to placing an answer at every angle the question decomposes into. The audit form of this is layer 3 in our 22-item checklist.
For link building, treat it as a widening of what counts. Alongside links, unlinked mentions and placement in comparison articles and review sites now feed the judgment. Nothing was added to your workload — something that was previously hard to measure started to matter.
The ten steps
The steps depend on each other, and there is a reason the first two come first. You cannot improve what you cannot measure. Without a record of the state before you started, you will not be able to explain three months later what worked.
If you would rather check items in parallel than work in order, our 22-item checklist covers the same ground in the other form. How the services differ is covered in our comparison of AI search engines.
Add chatgpt.com, perplexity.ai and gemini.google.com to your GA4 channel definitions. Done when the AI channel shows as an independent number
Cover four types — recommendation, comparison, branded and how-to — and run each several times on each AI. Done when a baseline mention rate is on record
GPTBot (training) and OAI-SearchBot (search) serve different purposes. Done when the two are handled separately and the settings match your intent
Disable JavaScript in your browser and open the page. Done when body text, price and specifications are all visible without JS
This is the foundation for the ChatGPT and Copilot route. Done when key URLs are findable in Bing
AI treats the passage under a heading as the unit of extraction. Done when every H2 on your key pages is a question or a claim
State the time frame explicitly, as in "as of July 2026". Done when every paragraph containing a number carries a source
A study, a measurement or a case — numbers others would want to cite. Done when data only you hold is published as an article
A comparison article, a review site or a press release. Done when the placement is confirmable
Done when the same metric has been recorded two months running
Steps six through eight can run in parallel. Three to five are technical and six to nine are editorial, so if different people own them, run both tracks at once.
What you can skip
Before allocating effort, confirm what to leave out.
Google's guidance states that there is no special schema.org markup to add for generative AI search, that creating an llms.txt file is unnecessary, that content need not be chunked, and that no special writing style is required.
Controlled experiments point the same way. In an experiment Ahrefs published in June 2026, 1,885 pages that gained JSON-LD were compared against a control group of 4,000, and no significant increase in AI citations was observed. For llms.txt, a study of 137,000 domains found 97% of published files had never been requested.
Structured data is effective for rich results in search, and for product information it lets you verify consistency against your Merchant Center feed. It simply has not been shown to increase AI citations.
When and where results appear
Each layer runs on a different clock. Clearing a technical fault can change answers within weeks once pages are re-fetched, while accumulating third-party evaluation takes months. The same phrase covers work whose payback arrives at very different speeds.
There is an order to how it surfaces: impressions, then citations, then traffic, then conversions. The further downstream, the later it moves. Traffic that has not moved in three months does not mean nothing is happening upstream.
On instrumentation, Google announced generative AI performance reports in Search Console on 3 June 2026 — the first official data separating AI Overviews and AI Mode impressions from ordinary organic search. Only impressions are available for now, with no clicks, click-through rate or query data. Data begins on 18 May 2026 with no backfill.
Demand in Japan has already moved
If the internal verdict is "too early", make it after looking at the numbers.
In our survey of 400 people who had bought a home appliance, 29.5% had used generative AI before the purchase. Including those who said they considered using it or wished they had, the figure reaches 54.0%. One in three actually used it, and more than half had a reason to.
The traffic side shows the same thing. Reading our own GA4 as of July 2026, AI-referred sessions accounted for roughly 4.7% of the total — 377 sessions over the trailing 90 days, and accelerating on a 28-day basis. Against a market average cited at 0.1% to 1%, that is clearly ahead.
Demand has moved and the supply side has not caught up. While that gap remains open, the returns concentrate on whoever prepared.
Frequently asked questions
Start without changing your existing allocation. Add measurement (steps one and two) and the technical checks (steps three to five), which take little effort and almost no budget. Revisit the allocation once you have a baseline.
Yes. Unlinked mentions and placement in comparison articles now count as well. If your program targeted link acquisition alone, widen the target to placement itself.
No. Focused pages tend to be cited more readily than long comprehensive ones, and Google states that content need not be chunked either.
A full rewrite is unnecessary. Turning headings into questions or claims with the answer directly below, and attaching sources and dates to figures, is enough to change how extractable a page is.
Steps three and four. Checking robots.txt and removing JavaScript dependencies works at any size and costs nothing.
The technical work in steps three to five can usually be handled in-house. Outside help pays off on third-party placement (step nine) and on running the measurement continuously (step ten).
Summary
AI SEO is not a story about abandoning the SEO you have. As Google states, best practices remain effective, and past investment survives as a precondition.
Three things get added: making pages readable by AI crawlers, writing in a form an answer can use, and measuring how you appear inside AI. The first two of the ten steps are measurement because you cannot improve what you cannot measure.
Demand in Japan has already moved — 29.5% of appliance buyers used AI before purchasing, and 4.7% of our own traffic now arrives through it. As a moment to start, this is neither too late nor too early. We offer a free AI visibility assessment for retail and e-commerce businesses.
