How do I get started with generative engine optimization?
How we start a generative engine optimization program: crawler access first, a prompt baseline second, then page fixes, off-site sources and traffic measurement.
Most teams that ask how to get started with generative engine optimization expect a list of content tricks. We start somewhere less exciting, because the biggest early wins in a new GEO program tend to come from removing blockers, not from writing clever copy. This is the order we follow with new clients.
1. Let the right crawlers in
We check access before we check content. The robots.txt file, the CDN and firewall rules, bot protection and rendering all get reviewed. OpenAI documents different crawlers for different jobs: OAI-SearchBot supports ChatGPT's search features and GPTBot is separate, so the decision to allow or block each is a deliberate one. We also look at ChatGPT-User, ClaudeBot, Claude-User, Claude-SearchBot, PerplexityBot, Perplexity-User and Google-Agent.
The free tools are good for this step: the AI bots directory with 440+ crawlers, the robots.txt and llms.txt generators, and the Agentic Readiness Check.
Then we verify with data. In Promptwatch, Agent Analytics shows real-time, IP-verified crawler logs, so spoofed user agents are filtered out. It shows a citation rate per page and error tracking with failure explanations. On a first audit this is where we most often find something nobody knew about, such as a bot-protection rule blocking a legitimate crawler.
2. Build the prompt panel
We write 30 to 50 questions from three sources: what the sales team hears, what Search Console shows people searching, and what the engines suggest. The Prompt Explorer in Promptwatch pulls suggestions from Search Console, ranking keywords, People Also Ask or any page. Prompt Volumes adds search volumes and difficulty scores. We tag prompts by topic, intent and type, and set country and language.
Then we leave it alone for two weeks. The baseline is more valuable than early tinkering.
3. Study the fan-outs
Engines split a question into sub-searches. Promptwatch lists query fan-outs per model, and the GEO Inspector Chrome extension shows ChatGPT's fan-outs and the fetched versus cited sources, running locally. We use these to write the content brief: the sub-question a competitor answers and our client does not.
4. Fix the pages that are closest
We do not rewrite the whole site. We pick the five to ten pages nearest to being cited and improve them.
- Answer first. The answer sits in the opening, then the support.
- Specific and sourced. The original 2023 GEO research tested adding quotations, statistics and source citations as tactics. We use real facts and name the source, and we never invent numbers.
- Clear structure. Headings that match how people ask, short sections, tables where a comparison is real.
- Honest structured data. Google's guidance on AI features is that standard SEO fundamentals apply, and structured data must match the visible content.
- Current facts. Prices, features and dates are checked every time.
Content gap analysis with coverage scores shows where there is no page at all, and Content Refresh gives diff-style update suggestions for existing ones.
5. Work the off-site sources
Engines cite more than your site. Promptwatch's citation analytics shows page and domain level citations and offsite mentions, where the brand is named on third-party pages without a link. Reddit, YouTube and LinkedIn reports show the platforms engines borrow from. We use these to choose where to be present, and we stay honest there: no fake reviews, no undisclosed paid posts.
6. Measure beyond mentions
We connect visibility to visits with visitor analytics, which attributes AI-referred traffic and conversions through a lightweight script, the GTM template or the WordPress plugin. 360 Insights joins crawler logs, citation clicks and AI-referred visits per page, and Search Insights overlays Google Search Console and Bing Webmaster data.
The weekly rhythm
Each week we check Visibility Score and Share of Voice, compare periods with the A/B comparison, review the Action Board, where a Discovery agent suggests up to five evidence-backed actions per run, and choose two or three tasks. For Business and Enterprise clients, the Content Agent can plan, write and publish against gaps on a schedule. We use its review inbox, so nothing goes live unapproved.
Where to start
The free Explore plan gives you 10 prompts on ChatGPT only, enough to learn the interface. Essential at $95 a month includes all models with 50 prompts. Crawler logs are not in Essential, and start on Professional at $245 a month, which we recommend early because access is the first thing we check. Agencies can use Kick-off, Growth or Scale at $199, $399 and $799.
What we say up front
GEO does not come with guarantees, and the timeline varies. What you can do is remove blockers, improve the pages engines pick from, and measure each change against a baseline. We run client programs on Promptwatch because that whole workflow lives in one place.