What is ChatGPT saying about your company? Demandbase presents an agent for AI visibility

When a buyer wants to know which service provider fits their problem today, they increasingly no longer open a search engine but ask an AI assistant. The answer contains two or three names. Whether your company is among them, and what is said about you, you will not see unless you deliberately look. This problem has acquired a name: AEO, short for "Answer Engine Optimization". On 20 October 2026, the US vendor Demandbase presented an agent that is meant to monitor how AI assistants portray a company.

For mid-sized companies in Germany, Austria and Switzerland, the product itself is not the news. It is a US offering whose status and availability in DACH the press release does not state. The news is the question behind it: Does your business know what ChatGPT, Gemini and other assistants say about you? In this article, we put into context what Demandbase has presented according to its own statement, what is substantiated and what is not, what you can check independently of the vendor, and where a human should keep the final say with an agent like this.

What Demandbase presented on 20 October 2026

The source is a press release distributed via Business Wire and readable on Yahoo Finance ("Demandbase Launches New AI Agents to Lead the B2A Marketing Era", 20 October 2026, retrieved on 22 October 2026). The page itself labels it as a paid press release. It is therefore a statement by the company about itself, not independent reporting. The agents were presented at the company's own conference, GO, in New York.

Three interconnected agents for B2B marketing are named:

• AEO Agent: It is meant to track and influence how AI assistants portray a brand, down to the level of individual members of a buying group. According to the release, marketers can compare how major AI assistants describe and recommend their company, identify outdated information and content gaps, and then launch campaigns to improve positioning. It is also meant to detect anonymous, AI-triggered website traffic at account and contact level so that sales can approach prospects in a more targeted way.
• Campaign Agent: It links campaigns to revenue targets, identifies performance gaps, recommends adjustments and can coordinate with specialized agents.
• Audience Agent: It builds and maintains audiences based on ideal customer profiles, buying groups and account activity, identifies missing members of a buying group and keeps audiences ready for use.

As figures, the release cites "342 million contacts across 13 million companies". These are Demandbase's own statements about its data foundation, undated and without any indication of how they were collected. It further states that the agents are meant to reduce manual work "while marketers retain control" and to free up teams for strategic and creative decisions. Demandbase's chief executive, Gabe Rogol, is quoted with the sentence: "Marketing to people isn't enough anymore because buyers no longer start with a search." In free translation: addressing only people is no longer enough, because buyers no longer begin with a search.

The Gartner figure: what it says and what we could not verify

The release refers to a Gartner forecast according to which around 90 percent of B2B purchases will be intermediated by AI agents by 2028, with a spending volume of more than 15 trillion US dollars. The order of the sources matters: the figure does not come from Demandbase, which merely links to it. We have not read the linked Gartner release ourselves. A supplementary search turned up only third-party reports that attribute the figure to Gartner and associate it with an event in October 2025. We did not find Gartner's own version.

A forecast is not a measurement. Anyone who reads "90 percent" should remember: it describes what an analyst firm expects, not what is happening today.

We therefore adopt the figure only with this caveat: Demandbase refers to a Gartner forecast that we have not verified in the original. We would not use it as a justification for an investment. As an indication that buying behavior is shifting, it is usable, as long as it is read as an expectation.

Availability: announced, still open for you

This is the point that sales pages like to leave out. The press release describes the agents as "launched" and "introduced". Whether they are in a test phase, generally available or limited to certain regions is not stated. The release also mentions an architecture without its own user interface ("headless") with access via programming interfaces and the Model Context Protocol (MCP), but gives no information on availability in this regard.

• Status: "Launch" according to the release; a status such as "generally available" or "beta" is not given.
• Region: US company, presented in New York; regions for deployment are not named.
• DACH: unknown.

Accordingly, we treat the AEO Agent as an offering that is coming toward companies, not as something you can book and deploy today. The problem it describes, by contrast, exists regardless of whether you use this or another tool.

The problem behind the product: How are you portrayed in AI answers?

AI assistants answer based on what they find about your company: your website, directories, press releases, reviews, specialist articles, old announcements. This produces answers that are not always correct. An outdated price, a discontinued product, a wrong location or a description that makes a competitor look better: such errors go unnoticed because no human sees them in a report. The prospect receives the answer, makes a decision and never contacts you.

This is why the question of how you are portrayed in AI answers is real for mid-sized companies, even if you will never use a US tool. We have already reported on related figures on the blog, for example on the use of AI for local search. The principle remains the same: anyone who wants to be found by customers must check how the assistants portray them.

A clarification of terms helps. "GEO" (Generative Engine Optimization) and "AEO" (Answer Engine Optimization) are used inconsistently in practice. GEO often refers to optimizing content so that generative search systems cite it, while AEO places more emphasis on how assistants portray and recommend a company to individual users. Both describe the same basic problem: visibility in AI answers instead of in a list of search results. Demandbase calls its product "AEO Agent"; we consider the two terms interchangeable enough that you should not dwell on the label.

What you can check yourself without buying a product

An agent can take the work off your hands, but not the understanding. Start, therefore, with a simple sample check that an employee can complete in an hour:

1. Collect questions. Write down ten questions a prospect would ask: "Which providers of [your service] are there in [your region]?", "Roughly what does [your product] cost?", "Who is better, [you] or [competitor]?"
2. Ask the assistants. Put each question to the common assistants and record the answers in writing, with date and assistant.
3. Mark deviations. Where is your company not mentioned? Where do outdated or incorrect statements appear? Which sources are cited as evidence?
4. Look for causes. Usually the source of the error lies on a page you can change: your own website, an industry directory, an old press article.

This sample check is no substitute for ongoing monitoring, because answers change. But it shows whether the topic is urgent for you. If even the ten questions reveal gaps, a regular process is worthwhile.

Four questions for every provider of a visibility agent

If you receive offers of this kind in the coming months, four questions help you put them into context, regardless of the provider:

1. Which assistants are checked, and how often? Answers change. A one-off snapshot says little.
2. Where does the data come from, and where is it processed? Especially with visitor and contact data, the location of processing matters to you as the controller.
3. What may the agent do without your consent? Only observe and suggest, or also publish and approach prospects?
4. How does the provider demonstrate impact? Self-reported figures on the data foundation are not impact. Ask for references that you can verify yourself.

Which team at Amazing Agents is responsible: Nadja and Lotte

On amazing-agents.com, the topic belongs to the research team. Nadja is described there as the research lead. According to the website, her tasks include topic research, competitor scans, GEO visibility analyses and research briefings; she checks how a brand appears in AI-generated search results and identifies gaps that can be closed with targeted content. Lotte is the trend scout on the team: she spots weak signals and delivers a weekly trend report to Nadja, with a relevance rating and a recommended response such as adapt, watch or ignore.

The part of the Demandbase agent that detects anonymous website traffic for sales belongs, on our website, more to the outreach team with Kai and Michael. Caution is warranted here, however: anyone who assigns anonymous visitors to specific accounts or individuals is processing personal data. Whether and how this is permissible in Germany depends on the specific procedure and should be reviewed by your data protection advisers. The press release says nothing on this.

We do not claim that Nadja and Lotte do the same thing as the Demandbase agent. These are different offerings with different scope. What they have in common is the task: observe, put into context, name gaps. At our company, no research agent alone decides on publications. The website describes that in the content pipeline, Magdalena checks every draft for brand fit, tonality and formal correctness before it goes to approval, and that Natascha approves the content.

With and without an approval step: the difference for a visibility agent

Without an approval step, an agent that detects that an assistant is portraying your company incorrectly would independently generate content, launch campaigns and put sales staff on detected visitors. Errors would be costly and hard to contain: a text with a false claim about your offering, a campaign that does not fit your brand, an approach for which there is no legal basis. Because the sender is your company, all of these errors fall back on you.

With an approval step, the agent observes and makes suggestions. It shows where the gaps are and supplies drafts for corrections. A human checks whether the gap is real, whether the suggestion fits the brand and whether publication or outreach is permissible, and then approves. Demandbase states that marketers retain control; how exactly this step looks in the product is not described in the release. For our approach, the basic rule is: the agent prepares, the human decides.

What you should prepare now

Since the offering has only been announced and its status for DACH is unclear, preparation comes before use. Three steps are enough to start:

1. Run the sample check. Have the ten customer questions from above answered and document the results. This takes about an hour and shows how urgent the topic is.
2. Clean up your own sources. Check your website, your directory entries and your profiles for outdated information: services, prices, locations, contact persons. What is wrong there is often passed on by assistants.
3. Define responsibility. Decide who will approve in future when an agent suggests corrections or content, and who clarifies whether outreach based on visitor data is permissible.

How we verified what is in this article

The statements on the three agents, the data foundation, the quote and the availability status come from Demandbase's press release of 20 October 2026, which we read directly on 22 October 2026. We did not read the Gartner forecast in the original; it appears here only as a reference in the release. The statements on Nadja, Lotte and Magdalena come from amazing-agents.com. What is not stated in the release, we did not include: whether the agent runs as a beta or is generally available, what it costs and whether it is offered in Europe remains open. We have not tested the product.

The difference that matters

Whether your company appears in the answer of an AI assistant is not a question for the distant future. Which tool you use for this is secondary; what matters more is who checks and who decides. An agent without limits publishes and approaches prospects, even when it is wrong. An agent with limits and an approval step observes, shows you the gaps and the drafts, and waits for your okay before anything goes out under your name.