Which AI platforms should an AEO strategy target?

Published audience figures describe everyone who uses a product. They can't tell you whether the people who buy from you are among them, so we'd start with what your site has already recorded.

Two reports show visits that arrived from assistants. Google Analytics has an AI Assistant channel for visitors who come from sources such as ChatGPT, Gemini, Copilot and Grok. HubSpot's traffic sources report has an AI Referrals category, and clicking its session count breaks the visits out by platform. HubSpot's list includes ChatGPT, Claude, Perplexity, Microsoft Copilot, Google Gemini and Meta AI.

Read both reports with one limit in mind. They count recorded visits that started with a click on a link in an answer, so an answer that named your company without a click leaves nothing in them. A platform with few sessions can still be naming you in its answers.

Your CRM adds the commercial part. If contacts and deals keep their original source, you can see which assistant is recorded against the visitors who became customers. That source is the tracked visit, which isn't always where the buyer first heard of you. It's still a better guide than session counts alone, and we cover the setup in how to track traffic and referrals from AI answer engines.

Google's AI Overviews and AI Mode don't show up in either of those assistant reports. Google Analytics says its AI Assistant channel excludes them, and HubSpot keeps regular Google search under organic search. A click from an AI Overview arrives looking like any other organic visit.

The place to look is Search Console. Its generative AI performance report counts impressions, meaning the times a link to your site was shown in AI Overviews or AI Mode. It groups them by page, country, date and device. An impression means a link was displayed, which is a different count from a visit, so keep it in its own column beside the referral numbers.

The Gemini app is a separate product from those Search features. Analytics and HubSpot both treat Gemini as an assistant source, while the Search Console report lists AI Overviews and AI Mode. A plan that says "target Google" should say which of the two it means, because they're checked in different places.

Reports show where visits came from. They can't show what an assistant says about you. For that, take the questions buyers ask before they buy and put them to each platform you're considering. Write down whether your company is named, whether your pages are linked and which other sources appear.

A sample tells you how you appear for those questions on the day you ran them. It can't tell you how many buyers asked them, so read it beside the visit and CRM numbers.

Lean Labs' AEO service is a documented example of the method. Its service page describes building the question set from HubSpot sales calls, CRM insights, Search Console demand and competitor prompts. It then lists prompt tracking across ChatGPT, Claude, Gemini and Perplexity. Two things make that useful as a model. The questions start from the client's own sales and search records, and the same set runs on several platforms, so the results can be compared.

Suppose your HubSpot report shows 140 sessions from ChatGPT last quarter, 30 from Perplexity and 4 from Claude. Six of the contacts created from those visits reached a sales conversation, and five of the six have ChatGPT as their recorded source. Search Console shows AI impressions on your pricing page and two comparison pages. When you run twenty buyer questions, your company is named in several ChatGPT answers and in none of Perplexity's.

Those inputs put ChatGPT first, because recorded visits and sales conversations already come from it. Google's Search features come second. The impressions show your links are being displayed there, and you'd want to learn which pages earn them. Perplexity is a gap to investigate, starting with whether its crawler can fetch your pages. Claude stays on the watch list until its numbers change.

Each row is a requirement, and meeting it doesn't mean a page will be shown. Google's guide to generative AI features in Search says those features draw on its Search index. It also says indexing and serving aren't guaranteed, even for a page that meets every requirement.

The ChatGPT row has a detail that's easy to miss. OpenAI's crawler documentation describes OAI-SearchBot as the agent for search and GPTBot as the agent that collects training material, and it says each setting is independent. A site that blocked every OpenAI agent to stay out of training has also opted out of ChatGPT search answers, though OpenAI says it can still appear as a navigational link. You can allow the search agent and keep the training agent blocked. OpenAI publishes the IP ranges for the search agent and recommends allowing them too.

Perplexity's crawler page makes a similar split. PerplexityBot surfaces and links sites in Perplexity's search results, and Perplexity says it isn't used to crawl content for foundation models. A second agent, Perplexity-User, fetches a page when a person's question calls for it, and Perplexity says that fetcher generally ignores robots.txt. Perplexity also publishes the IP ranges its crawler uses and recommends allowing them. A firewall can turn the crawler away even when robots.txt allows it.

We'd check these settings first for each of the three, because a page has to be reachable by a product's search crawler before the writing on it can count there. Blocked search access is one of the errors we describe in the biggest AEO mistakes to avoid.

Claude, the Gemini app, Copilot and Meta AI are all on HubSpot's list of recognized sources, so a visit is labeled with the assistant's name when it arrives with that assistant as its recorded referrer. We'd add one to the target list when your records give a reason. That reason could be sessions that turn into contacts, or a buyer who mentions the assistant on a sales call.

Your market can supply a reason too. Suppose your customers are IT teams who spend the day in Microsoft 365. You'd have grounds to give Copilot the full set of buyer questions before its referral numbers show anything.

For the rest, run a short version of the question set. That gives you a starting reading to compare against when one of them starts sending visits.

Choosing platforms mostly changes what you check and measure. Google's guide says its AI features need no special files or markup and no separate style of writing for AI, and it treats the work as SEO. ChatGPT search and Perplexity fetch public pages through their own crawlers. So the writing is shared work. Clear answers to your buyers' questions are what you'd publish whichever platforms are on the list. We explain that work in how answer engine optimization works, and how it overlaps with search work in the difference between AEO and SEO.

The target list then decides where you verify access settings and read reports, and which platforms get the full set of buyer questions. A change in the referral or CRM numbers is a reason to revisit it, and so is a new assistant coming up on sales calls.