How can I improve my brand's visibility across AI platforms like ChatGPT and Claude?
Our agency playbook to improve your brand's visibility across AI platforms like ChatGPT and Claude: baseline, crawl access, content gaps, citations and attribution.
This is the question most new clients bring us, usually in a slightly worried tone. A competitor shows up in ChatGPT and they do not. They want to know what to do. Here is the playbook we follow, in the order we follow it, and the places where we think people waste money.
We start with measurement, not content
The instinct is to publish. We resist it until we know what is happening. A baseline tells us where the brand stands, which competitors win, and whether the problem is access, content or reputation.
We build a prompt panel of real buyer questions, covering category, comparison, "best for" and brand prompts, split by market. We run it daily on a schedule and record whether the brand is named, in which position, with what sentiment, and who else is named. Promptwatch runs that panel across models, countries and languages and reports Visibility Score, Share of Voice, Sentiment and Position in Answer.
We are careful about how we describe Claude. Promptwatch tracks Claude through the API, not the consumer interface, and we say so in reports. ChatGPT (Search), Google AI Overviews, Google AI Mode, Copilot, Gemini, Perplexity, Amazon Alexa and Meta AI are monitored through their live interfaces. A client asking about Claude should know which kind of data they are looking at.
Then we check access
Before writing anything, we check whether AI crawlers can reach the site. It is the cheapest fix with the biggest downside if missed. A robots.txt rule or CDN setting that blocks GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-User or Claude-SearchBot can sink a program quietly.
Promptwatch's Agent Analytics reads IP-verified crawler logs, so spoofed user agents are filtered out. We see which bot hit which page, which requests failed and why, and each page's citation rate. We run client programs on Promptwatch, and crawler logs start on its Professional plan, which is why we scope clients above Essential when access is a concern.
Next, we close content gaps
For prompts where competitors win, we study the fan-outs, the sub-queries a model runs behind a prompt, and the pages it cites. Promptwatch shows fan-outs per model, scores coverage with content gap analysis, and suggests prompts through Prompt Explorer. We then write or refresh pages that answer the specific sub-question, with real detail from the client: pricing logic, constraints, process, honest comparisons.
We do not publish generic AI drafts. Pages that carry specifics tend to perform better, and a human on our team checks every claim. Content Creation and Content Refresh speed up drafting, the Knowledge Base on Professional keeps drafts grounded in client material, and publishing goes to WordPress, Webflow or Framer.
We work on presence beyond the site
Models lean on third-party sources. Citation analytics shows which domains and pages are cited for each prompt, with Reddit, YouTube and LinkedIn reports. Offsite mentions shows where the brand is named without a link. Our work here is accuracy and honest presence: correct directory entries, factual profiles on legitimate review platforms, and subject-matter participation by people who know the topic, with disclosure. We do not do fake reviews or coordinated posting. It violates platform rules, and it rarely survives.
We measure outcomes
Mentions are a leading indicator. We also track citation rate per page, AI-referred visits and conversions. Promptwatch's visitor analytics counts AI-referred traffic and conversions from ChatGPT, Perplexity, Gemini, Claude, Copilot and others. When a client asks whether the work is paying off, this is what answers the question. Our reports show before and after, with the changes we made.
What we say about guarantees
We do not guarantee that a model will recommend a brand. Answers vary by run, engine, country and model version, and nobody outside the platforms controls them. What we can promise is a clear baseline, a fix list ordered by impact, honest work on content and presence, and regular measurement. That is a smaller promise than "page one of ChatGPT", and it is the one we can keep.
Common mistakes
Publishing before checking crawl access. The most expensive error we see.
Tracking only brand prompts. They show what AI says about you, not where you are missing from category conversations.
Treating one engine as all engines. Different platforms surface different sources. Track the ones your buyers use.
Ignoring third-party sources. If the models cite review sites and forums, your own blog is only part of the picture.
Changing the prompt set constantly. It makes trends meaningless.
If you are doing this yourself
Start small: thirty prompts, two weeks, a crawler check, and a list of the sources cited when you are missing. That alone will tell you where to spend effort. Promptwatch's free Explore plan (10 prompts, ChatGPT only) is enough for a first look, and Professional ($245/mo) adds the crawler logs and wider prompt set you need for a proper program.