AI search optimization: a step-by-step guide

What AI search optimization covers

This guide is the practical side of generative engine optimization. The GEO guide explains what the discipline is and how it differs from SEO; this one walks through the work in the order you would do it.

Every engine on the list writes an answer instead of returning ten links, but they do not all read the web the same way. That matters for the technical steps, because each one reaches your site through a different crawler or index:

• Google AI Overview and AI Mode use Google Search. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link, and that there are no extra requirements beyond that (Google Search Central).
• ChatGPT search relies on OpenAI's OAI-SearchBot. OpenAI says sites that opt out of it will not be shown in ChatGPT search answers (OpenAI).
• Perplexity indexes sites with PerplexityBot, which it says is used to surface and link websites in its results (Perplexity).
• Microsoft Copilot sends search queries to Bing when it grounds an answer in the web (Microsoft), so being in the Bing index is the entry point.
• Gemini is governed by Google's crawlers, with the separate Google-Extended token controlling whether your content may be used for Gemini training and grounding in Gemini Apps (Google).

AI search optimization vs LLM SEO

LLM SEO is another name for the same work. People also say AI SEO, GEO and, when the focus is on being the direct answer to a question, answer engine optimization. The tactics and the metrics are the same whichever label a team uses.

The name is slightly misleading. LLM stands for large language model, but most of what you can influence happens in the search step around the model: the engine runs a web search, reads the results and writes an answer from them. What a model learned in training changes slowly and is hard to move. What it retrieves today depends on pages that can be crawled, indexed and quoted, which is why the steps below lean on classic SEO foundations.

Step 1: Measure where you stand

Start with a baseline, otherwise you cannot tell whether anything you change later made a difference.

1. Pick the prompts your buyers ask. Write full questions, not keywords: category questions (best tool for X), comparisons (A vs B), alternatives (alternatives to A) and problem questions (how do I fix Y). Keep most of them unbranded, because a prompt with your name in it will nearly always mention you.
2. Choose engines and countries. Ask each prompt of every engine your buyers use, in each market you sell in. Answers can differ by engine and by location.
3. Record every answer. For each one, note whether your brand is mentioned, its position relative to other brands, whether your site is cited, how you are described, and which sources are cited.
4. Calculate the baseline. Mention rate (the share of answers that name you) and share of voice (your mentions as a share of all tracked brand mentions) per engine, plus your citation rate and a list of the most cited domains.
5. Repeat before you trust it. The same prompt can return different brands on different days, so collect several runs before treating any number as your starting point.

The AI visibility guide explains each metric and the measurement mistakes to avoid. Keep the list of cited domains: it becomes your target list in step 4.

Step 2: Make sure AI engines can reach your pages

No amount of content work helps if the engine cannot read the page. Check these before anything else:

• robots.txt. Make sure you are not blocking the crawlers that feed answers: Googlebot, Bingbot, OAI-SearchBot and PerplexityBot. A broad rule meant for training bots can remove you from AI search answers by accident. The free AI crawler checker shows which bots your robots.txt allows.
• Training vs search bots. Some bots collect training data while others power search answers. OpenAI documents GPTBot for training and OAI-SearchBot for ChatGPT search as separate bots, so you can block one and allow the other. Google-Extended does not affect inclusion in Google Search, according to Google.
• Firewalls and bot protection. A CDN or security rule that challenges unknown bots can block AI crawlers even when robots.txt allows them. Check your firewall logs for blocked requests from the bots you want.
• Content in the HTML. Put the important text in the page source rather than only in images or content loaded by scripts after the page renders.
• Indexing. Submit an up-to-date sitemap and check that key pages are indexed in both Google Search Console and Bing Webmaster Tools, since Google's AI features and Copilot both depend on a search index.

llms.txt is optional. It is a proposal from llmstxt.org for a Markdown file that points AI systems at your most useful pages. It is not an official standard, and Google states that you do not need to create new machine readable files or AI text files to appear in AI Overviews or AI Mode (Google Search Central). It is cheap to add, and some AI tools read it, so the llms.txt generator can draft one in a minute, but do not expect it to move your numbers on its own.

Step 3: Write pages engines can quote

Engines lift passages, not whole pages. Make the passages on your most important pages easy to lift and safe to repeat.

• Answer first. On your homepage, pricing, product and category pages, say what you are, who it is for, what it costs and how you compare within the first few sentences. The AEO guide covers answer-first writing in detail.
• Cover comparison and list intent. Buying prompts are mostly comparisons and "best X for Y" questions. Publish fair comparison pages and alternatives pages, and make sure you appear in the roundups and listicles that engines cite for those prompts.
• Publish something only you can. Original data, pricing tables, benchmarks you ran yourself, documented methods and real customer examples give an engine a reason to cite you rather than a page that repeats what everyone says.
• Keep your entity consistent. Use the same brand name, category description, pricing and key facts on your site, your social profiles, directory listings and review sites. Conflicting facts make you harder to describe with confidence.
• Use clear structure. One question per section, descriptive headings, short paragraphs, lists and tables, and an FAQ for follow-up questions. Add structured data such as Organization, Product or FAQPage where it is accurate.
• Keep key pages current. If a page states prices or features, update it when they change, so engines do not repeat stale facts about you.

Step 4: Get cited by the sources engines already cite

For buying questions, engines often recommend brands on the strength of other people's pages. The sources list from step 1 tells you which pages those are for your prompts. Work through it by type:

• Review sites. Claim and complete your profile on the review platforms that appear in your sources, and ask real customers for reviews. Never buy or fake reviews.
• Listicles and roundups. If "best X" articles keep getting cited, contact their authors with accurate information and a reason to include you. Some will not reply; the ones that do can shape many answers.
• Reddit and communities. Threads in forums and communities show up often for buying questions. Take part openly, as yourself, where you can actually help. Undisclosed promotion breaks community rules and tends to backfire.
• Wikis and directories. Keep directory and marketplace listings accurate. Wikipedia has strict rules on notability and on editing articles about your own company (Wikipedia: Conflict of interest), so earn coverage in independent sources rather than writing your own entry.
• Partners and integrations. Partner directories, integration pages and customer case studies on other sites are third-party confirmation of what you do.

Prioritize the domains that are cited across many of your prompts and several engines. One source that keeps reappearing is usually worth more than ten that show up once.

Step 5: Re-measure and iterate

Run the same prompts, on the same engines and countries, on a regular schedule, and compare against the baseline. Keep the prompt set stable: if you change prompts and pages at the same time, you cannot tell which change moved the numbers.

• Read per engine. An improvement in Perplexity and no change in Google AI Overview usually points to a difference in sources or indexing, not to your content as a whole.
• Watch sources as well as mentions. A new citation of your page, or a new appearance on a frequently cited third-party page, often comes before a change in mention rate.
• Judge trends, not screenshots. Engines that search at answer time can reflect a crawled change quickly, while what a model learned in training moves slowly. Give changes weeks, not days, before you judge them.
• Feed the gaps back in. Prompts where competitors appear and you do not are your next content and outreach targets.

What not to do

• Block every AI bot by default. A blanket disallow can remove you from ChatGPT search, Perplexity or Copilot answers along with the training crawlers you meant to stop.
• Hide text for AI only. Content shown to bots but not to people is cloaking, which search engines treat as spam.
• Stuff pages with prompts or keywords. Engines quote passages that read well and answer the question; repetition does not help.
• Manufacture third-party mentions. Fake reviews, sock-puppet forum posts and paid placements dressed up as independent opinion break platform rules and can damage how you are described.
• Optimize from a single answer. One screenshot is a sample. Decide from repeated measurements.

FAQ