AI search optimization is the work of getting your brand named and recommended when buyers ask AI search engines such as ChatGPT, Perplexity, Gemini or Google AI Overviews a question. At #10, we run it as a program with a fixed order. Make sure the engines can read the site, and record a baseline for the questions buyers ask. Fix the pages that should answer them, correct the facts other sites hold about you, then re-check the same questions. One quarter is enough to finish a first pass.
Most of what is written about AI search is a list of tactics. Lists are easy to agree with and hard to schedule. The question a marketing lead actually has is narrower: what do we do first, what can wait, and how will we know in three months whether any of it worked?
This guide answers that. We have covered the definitions and the measurement elsewhere, so this page stays on sequence. For background, start with our generative engine optimization guide. For the tracking setup, see our guide to measuring AI visibility.
What AI search optimization covers, and what it does not
AI search optimization sits on top of traditional SEO. Google says plainly that there are no additional requirements to appear in AI Overviews or AI Mode, and no special schema to add. A page has to be indexed and eligible to show a snippet, and the usual SEO best practices apply. Other AI-powered search tools work the same way in broad terms: the AI system searches an index, reads a handful of pages, and the AI model writes an answer from them.
So the program is not a separate discipline with its own secret checklist. It is a different way of choosing and judging the work. The questions are longer and more specific than in traditional search, and the search intent behind them is often clearer. The result you are after is a sentence in the AI answer that names you, not a position in the search results. And more of the answer comes from pages you do not own, such as review roundups, marketplace listings and forum threads.
What it does not cover: using generative AI tools to write more content, a guaranteed placement in any assistant, a single “AI score,” or a content volume target. Nobody outside the engines controls which sources they pick, and the answers change from one run to the next. The program improves your odds and measures what happens. Our GEO vs SEO comparison goes through what changes and what stays the same in more detail.
The order matters more than the checklist
The steps below overlap, but the order is deliberate. Each one depends on the step before it. Fixing pages before you have a baseline means you cannot tell what the fixes did. Pitching third-party sites before your own pages say the right thing means the pitch points at a weak page.
<figure class=“ar-fig”>
<svg viewBox=“0 0 640 330” role=“img” aria-labelledby=“aiso-plan-t aiso-plan-d”>
<title id=“aiso-plan-t”>A 12-week AI search optimization plan</title>
<desc id=“aiso-plan-d”>Six steps on a 12-week timeline. Access check in weeks 1 and 2. Questions and baseline in weeks 2 and 3. Assigning pages in weeks 3 and 4. Fixing pages in weeks 4 to 8. Fixing off-site facts in weeks 6 to 10. Re-checking and choosing next quarter in weeks 11 and 12.</desc>
<g style=“font-size:14px”>
<text x=“250” y=“28” style=“font-size:13px;fill:var(–muted)”>Week 1</text>
<text x=“370” y=“28” style=“font-size:13px;fill:var(–muted)”>4</text>
<text x=“496” y=“28” style=“font-size:13px;fill:var(–muted)”>8</text>
<text x=“600” y=“28” style=“font-size:13px;fill:var(–muted)”>12</text>
<line x1=“250” y1=“40” x2=“250” y2=“300” stroke=“var(–line)”/>
<line x1=“345” y1=“40” x2=“345” y2=“300” stroke=“var(–line)”/>
<line x1=“472” y1=“40” x2=“472” y2=“300” stroke=“var(–line)”/>
<line x1=“598” y1=“40” x2=“598” y2=“300” stroke=“var(–line)”/>
<text x=“0” y=“70” style=“font-size:15px”>1. Access check</text>
<rect x=“250” y=“54” width=“63” height=“24” rx=“5” fill=“var(–ar-c2)”/>
<text x=“0” y=“112” style=“font-size:15px”>2. Questions, baseline</text>
<rect x=“282” y=“96” width=“63” height=“24” rx=“5” fill=“var(–ar-c2)”/>
<text x=“0” y=“154” style=“font-size:15px”>3. Assign pages</text>
<rect x=“313” y=“138” width=“63” height=“24” rx=“5” fill=“var(–ar-c1)”/>
<text x=“0” y=“196” style=“font-size:15px”>4. Fix pages</text>
<rect x=“345” y=“180” width=“158” height=“24” rx=“5” fill=“var(–ar-c1)”/>
<text x=“0” y=“238” style=“font-size:15px”>5. Off-site facts</text>
<rect x=“408” y=“222” width=“158” height=“24” rx=“5” fill=“var(–ar-c3)”/>
<text x=“0” y=“280” style=“font-size:15px”>6. Re-check, decide</text>
<rect x=“566” y=“264” width=“63” height=“24” rx=“5” fill=“var(–ar-c2)”/>
</g>
</svg>
<figcaption>A first quarter in six steps. Blue steps are measurement, teal steps are work on your own pages, coral is work on other sites. The page fixes and the off-site work overlap because they use the same list of questions.</figcaption>
</figure>
Twelve weeks is a realistic first pass for a brand with one main category. A larger catalog or several markets stretches the page work, not the order.
Weeks 1 and 2: make sure the engines can read the site
Start with access, because every later step assumes it. You cannot optimize for AI search if the crawlers cannot reach the page. This is a short technical check, and it often turns up something.
- Crawler rules. Check robots.txt and your CDN or firewall settings for the crawlers that feed answers. OpenAI, for example, separates OAI-SearchBot, which decides whether a site can appear in ChatGPT search, from GPTBot, which collects training data. Each setting is independent. Blocking training is a reasonable choice for some brands. Blocking search by accident is not.
- Bot protection. Security tools sometimes challenge or block crawlers that are not Googlebot. If the site returns a challenge page, an assistant has nothing to read.
- Content in text. Answers that live only in images, PDFs behind forms, or tabs that load after a click are harder to read and quote. Prices and specifications belong in the page’s text, and so do sizes and policies.
- Snippet controls. A
nosnippetor tightmax-snippetsetting left over from an old project will keep Google from quoting the page. - Search Console and Bing Webmaster Tools. Verify both, for every search engine your buyers use. Bing’s index also feeds Microsoft Copilot, and Search Console is where you will see whether the pages you fix are indexed.
None of this is glamorous. It is also the cheapest step, and skipping it can make the rest of the quarter worthless.
Weeks 2 and 3: write down the questions and take a baseline
Next, agree the questions you want to be part of the answer for, and record what the engines say today. This list drives every later decision, so it is worth an argument with sales and support before it is fixed.
Build it from evidence, not from what the team hopes buyers ask. Search Console queries with impressions, paid search terms with real volume, site search logs, and the questions your sales or service team hears every week are all good sources. Group the questions by what the buyer is trying to do:
- Problem questions, where the buyer has a need but no brand in mind.
- Shopping or shortlist questions, where they are choosing what or whom to buy.
- Comparison and brand questions, where a name is already in the question.
Then run each question on the engines that matter to your buyers, more than once. Answers vary between runs. Record whether your brand is named, whether it is recommended, which sources are cited, and which competitors appear. Write the date and the settings down.
We learned two things about this step the hard way. For a premium California olive oil and wine brand we track monthly, the first query set was partly hand-picked, and some of the questions had almost no search demand behind them. When the client asked how success would be measured, we rebuilt the set around queries backed by Search Console impressions and keyword volume, and kept the original questions as a separate group so the trend stayed comparable. We also found that two runs an hour apart disagreed on several questions. Since then each question runs several times, and an empty or failed answer is logged as missing rather than counted as “not named.”
Our sample AI visibility report in Labs shows what the output looks like: prompts grouped into sets, results per assistant, cited sources, and the actions that come out of it. The brand and numbers on that page are invented. The structure is the one we use. For a full method, see our guide to AI search visibility audits.
Weeks 3 and 4: decide which page should answer each question
Every question on the list should have one page on your site that is meant to answer it. When nothing owns a question, an assistant has nothing of yours to cite. When three pages half-answer it, they compete with each other and none of them is clearly the best source.
| Question type | Page that should answer it | What it usually needs |
|---|---|---|
| Problem question | A guide or article | A direct answer in the first paragraph, then the depth |
| Shopping or shortlist | Collection, category or service page | Who it is for, what it costs, how it compares |
| Comparison | A comparison page or table | Named options, the same attributes for each, units |
| Brand question | About, product or FAQ page | Plain facts: what you sell, where, since when, for whom |
This step is also where you protect what already works. If a page earns meaningful search traffic, improve it in place. Do not rewrite it from scratch or move its URL to fit a new plan. When two pages compete for the same question, merging the weaker one into the stronger with a 301 redirect usually helps both classic rankings and the odds of being cited.
Weeks 4 to 8: fix the pages closest to the buying decision
This is the longest step and where most of the effort goes. Work through the assigned pages in order of commercial value, not in order of how easy they are to edit.
The changes are mostly structural. Good content for AI search engines is good content for people, organized so a single passage can be quoted. Open the page with a plain answer to the question it owns, written so it still makes sense if it is lifted out on its own. Put comparisons and specifications in tables with units. Add a short FAQ in the buyer’s own words, and mark it up with structured data that matches the visible text. State the facts an assistant needs to describe you correctly: what you sell, who it is for, where you operate, what it costs or how pricing works.
We go through these patterns with before-and-after examples in our guide to answer engine optimization, so we will not repeat them here.
The best evidence we have for this step is a retrofit. For an industrial LED distributor, we added citable lead paragraphs, tables of specifications and comparisons with units, and FAQ blocks with FAQPage and TechArticle schema across an existing library of technical articles. We did it without rewriting the core content and without changing URLs, so the pages kept the search history they had already earned. That is the default we recommend: restructure what you have before you write anything new.
New pages come last, and only for questions on the list that nothing on the site can answer.
Weeks 6 to 10: fix the facts other sites hold about you
Assistants quote pages you do not control. For shopping questions in particular, the cited sources are often review roundups, marketplace listings and directories rather than brand sites. If those pages leave you out, or carry last year’s price, the answer will too.
Work from the cited sources you recorded in the baseline:
- Feeds and profiles you own elsewhere. Google’s guidance on its AI features specifically mentions keeping Merchant Center and Business Profile information current. The Business Profile also feeds local search, so the same facts show up in maps and in AI responses. Do the same for marketplace listings, wholesale directories and any industry databases that list you.
- Pages that cite your category but not you. A buying guide or roundup that answers your shopping questions and names your competitors is worth an approach. Give the editor something real to work with: samples, test results, a clear explanation of how you differ.
- Wrong facts. If an assistant repeats an old price, a discontinued product or the wrong location, find where it came from and correct it at the source.
This work starts while the page fixes are still under way, because it uses the same list of questions. It also takes longer to show up. A roundup editor may update once a year.
Weeks 11 and 12: re-check, read the trend, choose next quarter
Run the same questions again, on the same engines, with the same settings. Compare the groups, not individual answers. One answer flipping from “named” to “not named” is noise. A whole group of shopping questions moving is a signal.
Keep four things separate in the report:
- Mentions and recommendations in the AI search results you sampled.
- AI citations, meaning links to your pages as sources.
- AI referral traffic in analytics, by assistant, and where those visitors land.
- Orders, leads or inquiries from that traffic.
Mixing them hides what happened. A brand can be mentioned more often while sending no more visitors, or receive more AI referrals from an answer that never cited it.
The same olive oil and wine brand shows why the groups matter. Its first baseline found the brand cited on most of its health and how-to questions, and every one of those citations pointed to an educational blog article. On the shopping questions it was cited on none. Meanwhile, the visitors who did arrive from AI assistants mostly landed on product pages. That told us where the next quarter’s work belonged: the collection and product pages, not more articles.
Close the quarter by choosing the next set of pages and sources, based on what moved and what did not. Visibility in AI answers compounds slowly, the same way search rankings do, so the second quarter usually builds on the first rather than starting over.
What to skip this quarter
Some work feels productive and does little. Most of it is content optimization aimed at the tool rather than the buyer. These are the requests we most often push back on in a first quarter.
| Tempting | Why it can wait |
|---|---|
| Publishing many new AI-written articles | Generic text sounds like the rest of the web and gives an assistant no reason to pick you |
| Adding every schema type available | Markup helps machines read a page. It is not a result, and Google says no special schema is needed for AI features |
| An llms.txt file as the main fix | Reasonable to add, and we publish one, but Google says AI-specific files are not required. It does not replace the pages |
| Rewriting pages that already rank | Risks the traffic you have for a gain you cannot yet measure |
| Tracking one “AI visibility score” | A single number hides which questions and which engines moved |
Our own experience backs this up. We built our citation tracking on nr10.com before we had much to show for it, and the first baseline was zero citations. Schema was in place. The pages were not yet the best answer. The fix was content and authority, not more markup.
Questions brands ask about AI search optimization
What does an AI search optimization agency actually do?
It checks how assistants describe your brand today, finds the questions where you are missing or wrong, and fixes the pages and outside sources behind those answers. Then it re-runs the same questions on a schedule. Expect a baseline, a ranked list of page changes, off-site work, and a report that keeps mentions and citations apart from traffic.
How long before we see results from AI search optimization?
Access fixes and page changes can show up within weeks for engines that search the live web, once the pages are recrawled. Changes that depend on third-party sites take longer. Read the trend over two or three monthly checks rather than judging a single run.
Is AI search optimization different from SEO?
It builds on the same foundation: crawlable, indexed, useful pages with clear structure. The differences are in what you measure and where you work. You track whether you are named in answers, and you spend more effort on outside sources and on passages that can be quoted alone. Our GEO vs SEO guide covers this in full.
Can you guarantee our brand will show up in ChatGPT or AI Overviews?
No. The engines do not publish how they pick sources, and their answers change from run to run. We can improve the odds and measure the change honestly. Anyone offering a guaranteed placement is selling something they do not control.
What should an ecommerce brand fix first for AI search?
Usually the collection and product pages behind its shopping questions, plus the product feeds and marketplace listings assistants read. Educational content tends to earn citations first. Shopping answers are often where the brand is missing and where the buyer is closest to purchase.
Start with a baseline you can repeat
The first useful step is small: a written list of your buyers’ questions, run on the assistants they use, with the results recorded and dated. Everything else in this guide is ordered from that list.
If you want us to run it, our AI search optimization work starts with that audit and turns it into the quarter’s plan. It sits alongside our technical SEO audits and ongoing SEO programs. To see the report format first, open the AI visibility sample in Labs, then talk with #10 about the questions your buyers are asking.
