How to rank on AI search results: the process we run for clients
How to rank on AI search results, from an agency that does it for clients: prompts, crawler access, content, third-party sources and measurement, step by step.
Clients ask us how to "rank" in ChatGPT as if there were a page two. There is not. AI engines write an answer, name a few brands and cite a few pages. So we reframe the goal in the first meeting: be named and cited for the questions your buyers ask, and make sure the visit that follows converts. Then we follow a process, and the measurement side runs on Promptwatch.
1. Prompts before pages
We never start with content. We start with the questions. In a workshop with the client we collect 30 to 50 buyer prompts from sales calls, support tickets, Search Console and People Also Ask, then label each as organic, brand or competitor comparison. Promptwatch's Prompt Explorer speeds this up, and persona stacking lets us run each question as different buyers would phrase it.
Then we look at query fan-outs, the sub-queries an engine runs before answering. They show what the engine actually wants to know, and they frequently reveal an angle the client's pages do not cover.
2. Check access first
Before writing anything, we check whether engines can read the site. This is often the cheapest win. We look at robots.txt, bot protection rules, JavaScript rendering and server errors for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and the other real bots. Promptwatch's Agent Analytics shows IP-verified requests in real time, with errors and explanations, so we can see which pages fail and why. Crawler logs start on the Professional plan, which we flag in the scope. The weekly site audit adds crawlability checks.
If a client's pages are not fetched, we fix that before any content work. Otherwise the content is written for nobody.
3. Write for the sub-questions
When we do produce content, we write pages that answer the main question in the first few lines, then cover each fan-out question under its own heading. We add specifics a reader can use: named options, steps, real figures with a source, constraints. We keep the page current and avoid padding.
For existing pages, the Content Refresh feature suggests diff-style updates, and content gap analysis scores coverage against the prompt panel. For new pages on Business plans, the Content Agent can plan and draft against gaps with a review inbox. We always keep the review step on, so a person at the client or on our team reads before publication.
4. Work the sources outside the client's site
This is the step clients underestimate. Engines frequently cite review sites, comparison articles, Reddit threads, YouTube videos and LinkedIn posts. If those do not mention the client, a better homepage will not fix it. Promptwatch's citation analytics list cited pages and domains for each prompt, and the Reddit, YouTube and LinkedIn reports show community sources. Offsite mentions show where the client is named without a link, and Citation Trends shows how long a source stays cited.
We advise clients to participate honestly: give accurate information to publishers, answer questions openly and correct what is wrong. We steer away from anything that looks like manufactured mentions. It does not hold up, and a damaged reputation can end up in the answers.
5. Measure what happens next
Three things get measured every month:
- Presence. Visibility Score, Share of Voice, Sentiment and Position in Answer across the fixed panel, by engine and country.
- Citations. Which of the client's pages are cited, for which prompts, and which competitors are cited instead.
- Traffic. AI-referred visits and conversions per engine from visitor analytics, installed through a small script, the GTM template or the WordPress plugin.
360 Insights joins crawler logs, citation clicks and AI-referred visits per page, which helps us decide whether a page needs content, access or promotion.
6. Keep a fixed panel and change one thing at a time
AI answers vary between runs, countries and personas. We freeze the panel for a quarter, review trends weekly and monthly, and change one lever at a time so we can say what worked. The Action Board turns the data into a prioritized list, and its Discovery agent runs weekly and suggests evidence-backed actions.
What we tell clients
We do not promise a position. Nobody can, because the engines choose their sources, and they change how. We promise a clean method, measured results and a ranked list of next actions. Where Google's own guidance matters, such as how its AI features choose pages, we point clients to Google's Search Central documentation instead of paraphrasing it, because that is a policy question and not an agency opinion.
Where other tools help
We also use or have tested other tools. Mentionable and Ceyo track daily at a lower price. AthenaHQ and Scrunch AI are platform-style products at higher entry prices. Semrush and Ahrefs have AI modules for teams who live there. For monitoring alone they can be a good fit. For the process above, the crawler logs and visitor analytics are what we could not do without, which is why our programs run on Promptwatch.
If you are starting on your own, do steps one and two this week: write the prompts and check crawler access. They cost little and decide whether anything else works.