# How to market your AI Agent

> You built your AI agent, congratulations! Now let's get it recommended when people ask AI what to use.

- Author: Ash Gaikwad (https://ashgaikwad.com)
- Published: 2024-11-25
- Topics: marketing, ai agent, geo, seo
- Canonical: https://ashgaikwad.com/blogs/how-to-market-your-ai-agent

Changing times for everyone. In one night people can get a prototype ready and within a week they are ready to ship the
first version. Creating software products is easy nowadays. Many products are created, but very few manage to get real
users once they launch.

So let's assume you are done with your first week. The agent works, the landing page is live and you have told your
friends about it. This article will help you with the next part, which is getting your first real users (early adopters,
the people who try new things before everyone else). And I want to focus on the channel that changed the most in the last
few years, which is people asking an AI assistant what they should use.

## Your next user will probably ask an AI first

A few years ago you typed four words into a search engine and opened five blue links. Now many people type a full
question into an AI assistant, something like "what is a good AI agent that can answer customer emails for a small
online store?", and they get one answer that names a few products and explains the trade-offs.

If your agent is not in that answer, the person never learns you exist. They do not scroll to page two because there is
no page two. That is the reason a new discipline showed up next to search engine optimization (SEO). It is called
generative engine optimization, or GEO.

## What GEO means in plain words

SEO asks, "How visible is my product when people search the web?" GEO asks, "How visible is my product when people ask
an AI for advice?"

The term comes from a research paper called
[GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735), published in late 2023 by researchers from
Princeton University and other institutions, and presented at the KDD conference in 2024. They tested
different ways of rewriting web content and measured how often that content showed up in AI-generated answers. Some
changes improved visibility by up to 40%. The methods that helped most were adding citations to credible sources, adding
quotations and adding concrete statistics. Old tricks like keyword stuffing did not help.

You will also hear it called answer engine optimization (AEO). Some people say it is all still SEO, and they have a
point, because good SEO habits help GEO a lot. The difference is in what you measure. In SEO the unit is the keyword and the ranking. In GEO the unit is the question and the answer. You care whether
the AI mentioned you, whether it recommended you, where it placed you next to competitors, what it said about you and
which sources it relied on.

## How an AI assistant decides what to mention

An AI assistant answers from two places. The first is what it learned during training, which is a large snapshot of the
public web from some point in the past. The second is live retrieval, where the assistant searches the web while it
answers, reads a handful of pages and then writes a summary. Most assistants now do both.

This has two consequences for you. For training data, the question is whether your product was written about in enough
places, clearly enough, for the model to remember what it is. For live retrieval, the question is whether the pages it
finds are easy to read and actually answer the question. In both cases the AI leans heavily on what other people say
about you (reviews, comparisons, forum threads, articles) and not only on your own homepage.

So GEO comes down to three jobs. Be easy to find. Be easy to understand. Be talked about in the right places.

## Step 1: Write down the questions your users ask

Everything starts with the questions. Open a document and write 20 to 50 questions that a potential user might ask an AI
before they would find your agent. Write them the way a real person would type them, in full sentences.

Cover different moments of the buying process. Some questions are about the category ("what are the best AI agents for
bookkeeping?"). Some are about a problem ("how can I stop missing leads that come in after office hours?"). Some are
about a specific situation ("which AI assistant works well for a two-person law firm in Canada?"). Some compare you with
others ("what are alternatives to [a competitor]?"). And some are about details that decide a purchase, like price,
integrations, languages or data privacy.

If you already have users, ask them what they searched for before they found you. Their exact words are worth more than
anything you invent at your desk.

## Step 2: Measure where you stand today

Now ask those questions to a few different AI assistants your audience uses, in a fresh chat each time so your own
history does not influence the result.

AI answers are not deterministic. The same question can give a different answer an hour later. So ask each question more
than once, and think in percentages rather than single results. "We were mentioned in 6 out of 30 answers" tells you far
more than one screenshot.

A simple spreadsheet is enough. For every answer, record the following.

1. Whether your agent was mentioned at all.
2. Whether it was recommended, which is different from being mentioned in passing.
3. Its position compared to the other products in the answer.
4. Which competitors appeared.
5. How the AI described your agent, in its own words.
6. Which sources or links the AI cited.

The last two columns are the most interesting. The description tells you what the AI believes your product is, and the
sources tell you where that belief came from. If a competitor keeps getting recommended and the AI keeps citing the same
three review pages, you now know where you need to show up.

Do this once as a baseline and write down the date. You will come back to it.

## Step 3: Make your website easy for machines to read

Many AI crawlers do not run JavaScript the way a browser does. If your landing page only shows content after a big
script loads, the crawler may see an almost empty page. Make sure your important pages send real text in the HTML, which
is called server-side rendering or static generation. You can check this quickly by viewing the page source in your
browser. If you cannot find your headline and description in there, neither can a crawler.

Next, check your `robots.txt` file. Some hosting setups and security settings block AI crawlers by default. Decide
consciously whether you want them in, and if you want to be recommended, the answer is usually yes.

Then give the machines the facts in a structured way. Add [structured data](https://schema.org/) to your pages, such as
`SoftwareApplication` or `Product` for your agent, `Organization` for your company and `FAQPage` for your questions and
answers. It does not guarantee anything, but it removes guesswork about what your page is.

Finally, make sure there is one clear page that says, without marketing fog, what the agent does, who it is for, what
it costs, what it integrates with and what it cannot do. When an AI tries to describe your product, this is the page you
want it to read. You can also publish an `llms.txt` file, a proposed convention for giving AI systems a plain summary of
your site. It is cheap to add, but do not expect miracles from it alone.

## Step 4: Write content that deserves to be quoted

An AI assistant quotes pages that answer the question well. So write pages that answer your users' questions from Step 1
directly and honestly.

Put the answer near the top of the page in plain language, then explain the details. Write so that any paragraph makes
sense on its own, because the AI might lift only that paragraph, and use headings that match the questions people ask.

Be specific. "Our agent saves time" says nothing. "Our agent drafts replies to about 70% of incoming support emails, and
a human approves them before they go out" is something an AI can repeat. Only use numbers you can stand behind, and link
to where they came from. The GEO research found that citations, quotes and statistics made the biggest difference,
which matches common sense. Specific, sourced content is more useful to a reader and more trustworthy to a machine.

Write the pages your competitors skipped, such as honest comparisons ("us vs. them, and when you should pick them"),
guides to the problem your agent solves and clear pricing explanations. Being fair to competitors makes your page more
credible, and credible pages get cited.

Keep things fresh. Show a visible "last updated" date on important pages and actually update them. AI assistants that
search live tend to prefer recent information, especially in a fast-moving field like AI agents.

And please do not stuff keywords or publish a hundred thin pages written by a model in one afternoon. It did not work
for search engines and it does not work for AI either (I say this as someone whose site is written, or more like typed,
with help from AI).

## Step 5: Get talked about in other places

This is the step most builders skip, and it is often the one that matters most. When you look at the sources column from
Step 2, you will probably notice that AI assistants cite third-party pages far more than company homepages. That makes
sense. An independent review is more convincing than a company praising itself, for humans and for machines.

So go where your users already are. List your agent in the directories for your category and ask happy users for
honest reviews on sites your audience trusts. Answer questions in community forums where people ask for
recommendations, and be open that you built the product. Write guest articles, join podcasts and get listed in your
partners' documentation. Even mentions without a link seem to help, because they teach the AI that your name belongs
with a certain topic.

Consistency matters as much as volume. Describe your agent in the same words everywhere, with the same name, the same
one-sentence description and the same category. If your website says "AI receptionist", your directory listing says
"chatbot platform" and your social profile says "automation tool", the AI has to guess, and it may guess wrong.

## Step 6: Measure again, then repeat

GEO is not a one-time project. It is a loop. You observe, you measure, you find the gap, you fix it and you measure
again.

Every month or so, run the same questions against the same assistants and compare the numbers with your baseline. Check
whether your mention rate went up, whether the AI describes you the way you want and whether new sources cite you. Keep
the question set mostly stable so the comparison is fair.

Be patient. Live search results can change within days, but what a model learned during training only changes with a
new model version, which can take months.

## Mistakes I see often

The first mistake is treating GEO as a trick. There is no secret prompt or hidden tag that makes an AI love your product.
The foundation is the same as it always was, which is a useful product and clear, honest information about it.

The second mistake is looking only at your own website. Your homepage is one voice. The AI listens to the whole room.

The third mistake is chasing a single score. A number going from 42 to 57 means nothing if the people asking those
questions would never buy from you. Measure the questions that matter to your business and care about whether the right
people find you.

## A plan for your first week

If you want to start today, here is a small plan that fits in one week of evenings.

1. On day one, write down 20 questions your users would ask an AI, in their words.
2. On day two, ask those questions to three different AI assistants, twice each, and record the results in a
   spreadsheet.
3. On day three, view the source of your main pages, check your `robots.txt` and fix anything that hides your content.
4. On day four, write or rewrite one clear page that says what your agent does, who it is for, what it costs and what it
   cannot do, and add structured data to it.
5. On day five, write one useful article that answers the most common question on your list better than anything the AI
   cited.
6. On day six, list your agent in two relevant directories and ask three happy users for an honest review.
7. On day seven, rest. Then put a reminder in your calendar to measure again next month.

That is it. None of this needs a big budget or a marketing team. It needs you to understand your users' questions and to
make sure the answer to those questions includes you. Good luck with your launch, and I would love to hear what you
build.