Answer Engine Optimization (AEO), Part One: What It Is and How to Test It

The most popular use of AI is to get expert answers. As a local business owner it may be difficult to get cited by AI on topic expertise alone, but it is still important to understand what AI can discover about your business, your local competitors, and local services and products.

It is also important to understand that AEO cannot really be measured. In the world of Answer Engines there is no such thing as “page rankings” or any reliable metric of performance. AI remembers some information about each user, intentionally adds some flavor (aka temperature) to keep answers fresh, and considers the entire discussion thread when generating a response. Combined, these variables create too much variation to measure.

Testing AEO

Given that limitation, conducting AEO research on your business is still useful to identify what AI knows about you, your competitors, and opportunities for enhancing your messaging.

Although there are automated services for testing AEO, the best approach is to do live tests using humans who understand your business. As a business owner you should probably be involved at least occasionally to catch subtle mistakes. If you work with a trusted marketing team they may do the initial review and bring any unclear points to your attention.

Before elaborating on the stages, let’s consider how to conduct this research.

AI Platforms

It is time consuming to do a complete review for each AI platform. In 2026 Google’s Search AI has proven very effective and we recommend using that as your baseline for all tests. It also gives you some insight into Google Assistant, which shares the same foundation.

With your most important prompts, such as the competitor and customer-sentiment prompts we’ll cover in Part Two, test in ChatGPT and Claude as well. Between them, you also get indirect coverage of the major voice assistants. (Perplexity may be interesting too because its core model is DeepSeek; use the Sonar model if you test with it.)

Unbiased Research

If you are logged into a Google account then Google’s Search AI knows a lot about you when answering questions. Claude and ChatGPT (or any model) may track some key data about you too. Ideally you do not want your AEO research to be biased by this collected data. The best solution is to use an incognito browser window, and to NOT log in to a Google account in the incognito session. Google Search AI will still know your location but most of your personal data will be unavailable.

Every stage of this research assumes an incognito window. Some stages also call for a fresh window for every single prompt; we’ll flag those as we go.

If you are researching your company but are traveling (or geographically distant) from that business then prompts about services or products may need more precise language like “local to Madison WI”. Even then, AI may get the wrong location. If it does, don’t sweat it, just share your prompt with someone local to the business to try on your behalf.

Session bleeding (or drift) is the tendency for answers to take previous answers in the same chat into account. Generally that’s helpful, since it narrows you in on a solution, but it’s exactly what we don’t want in most of this research. When a stage calls for a clean context, close the incognito window after each answer and start fresh for each new prompt.

In Part Two we’ll walk through the six research stages themselves, from checking what AI believes about your identity to hearing what it thinks your customers are saying.