The Marketing to AI Agents Handbook

My first online purchase was in 2010, almost 16 years ago. I was a student in Pakistan and I had saved for months to buy a smartphone. Before paying anything, I spent weeks going through store pages, reviews and forums, trying to work out if the shop was even real. And after all that, I still paid cash on delivery, because I just didn't trust a website with my card.

These days I ask my AI assistant to reorder the toilet roll, find me a vacuum cleaner under $50, or get bedsheets like my last pair, and it just does it. It checks the stores, reads the reviews and comes back in minutes.

That is a big shift, and honestly most of our marketing isn't ready for it. We spent years learning how to impress people. Now there is someone else doing the shopping, and it doesn't care much about your hero image or your clever tagline. It cares about your specs, your reviews and your return policy.

So this handbook is how I think about marketing to that new customer. How agents shop, what to fix on your site, how to get the proof and mentions they look for, and how to check if you're ready.

What is marketing to AI agents? It is making your product easy for AI assistants to find, understand, trust and recommend when they research or buy on someone's behalf.

TL;DR

1. Fix your product pages first. If an agent can't find your price, specs, shipping and returns, it moves on to someone else. Start with your top 10 pages.
2. Get real reviews, early. Ask your first 50 customers, and put the reviews on your product pages where agents can actually read them.
3. Get people talking about you. A Reddit thread, a comparison article, a newsletter mention. What others say about you counts more than what you say.
4. Check your robots.txt today. It takes 10 minutes, and you want to be sure you aren't blocking the AI crawlers by accident.
5. Ask the assistants about your category once a month. ChatGPT, Claude, Gemini, Perplexity. See if you come up, and what they say when you do.

Contents

OKARA TERMINAL — AGENT READINESS SCAN
> loading handbook...
> 9 chapters found

>01  Your next customer is an AI agent — they're already shopping
>02  How agents shop — specs, not slogans
>03  Brand still matters — just differently
>04  Make your product readable — if an agent has to guess, it moves on
>05  Proof, not promises — reviews, policies and certificates
>06  Get mentioned — what others say about you counts more
>07  Let the agents in — robots.txt, checkout and your stance
>08  Who owns the agent — and what stops working
>09  Are you ready? — a five-step check you can do this afternoon

01 — Your next customer is an AI agent

This is not a someday thing. It is already showing up in the numbers.

Adobe tracks more than a trillion visits to US retail sites, and in September 2026 traffic coming from AI sources was up 71% from a year before. The bigger story is how those visitors behave. They converted 90% better than everyone else and spent 83% more per visit [1].

Which makes sense when you think about it. By the time someone clicks through from an assistant, the agent has already done the comparing, the review reading and the "is this shop even legit" part. The click is the last step, not the first.

The checkout is moving into the chat as well. Google's Universal Commerce Protocol lets people buy from eligible US retailers right inside AI Mode and the Gemini app [2]. Shopify now lets millions of merchants sell to shoppers in ChatGPT, Copilot, AI Mode and Gemini [3].

So we are marketing to two customers at once now. The person, and the agent working for them. And the second one reads everything, compares everything, and has zero patience for fluff.

02 — How agents shop

Humans are emotional, we buy into a story. A nice founder video, a friend who swears by something, a limited drop, and we get FOMO'd into it. Agents don't work like that. They are logical and very, very diligent, so a lot of the tricks that work on us just don't work on them.

Here is what matters most.

Specs over slogans. "Luxuriously soft" means nothing to an agent. "100% cotton, 400 thread count" does, because it can match that to what I asked for and compare it across ten stores.

No impulse buys. An agent sent for eggs doesn't come back with a chocolate bar. All the upsells, "customers also bought" rows and countdown timers we spent years optimizing are mostly invisible to it. The impulse is gone.

Reviews matter again. Agents can analyse thousands of reviews in seconds. Heck, they can smell fake ones too, like 200 five-star reviews that all landed in the same week saying the same thing.

So what does that mean for your product page? Mostly, turning the vague stuff into facts. A few examples:

• "Luxuriously soft sheets" becomes "100% cotton, 400 thread count, fits mattresses up to 16 inches deep."
• "Lightning-fast shipping" becomes "Ships in 1 business day, arrives in 2 to 4 days in the US."
• "Hassle-free returns" becomes "30-day returns with a free return label, refund within 5 business days."
• "Built to last" becomes "2-year warranty, and replacement parts are sold separately."

You don't lose the human reader by doing this. People like facts too, they're just more forgiving when you leave them out.

03 — Brand still matters, just differently

So does brand still matter when agents do the buying? Yes, but it's a bit more nuanced now. Here is how I think about it.

If you tell your agent, "get me a Patagonia jacket," the brand has already won. The agent's job is just to find your size at a decent price and get it to you on time. All the brand building Patagonia did over the years pays off right there.

If you didn't tell the brand name, the agent will find the options itself. Say "get me a warm jacket for a week in Iceland," and now nobody has won yet. It goes through dozens of products, reads the reviews, looks for discounts, checks who can deliver before you fly, and then picks.

That second kind of request is where most of us need to do the work. And there, branding moves away from polished ads towards information that is authentic and easy for a machine to read. What people say about you online starts to matter more than your best ad creative.

This is also why you now see GEO (generative engine optimization) sitting right next to SEO in marketing tools. SEO helps your pages rank. GEO helps your brand get named and cited inside the answers AI assistants give.

The rest of this handbook is about that second scenario: making sure the agent picks you even when nobody asked for you by name.

04 — Make your product readable

The rule is simple: if an agent has to guess, it moves on. It has plenty of other options and no reason to give you the benefit of the doubt.

Agents can read your product, but only what is actually there. Your page, your product feed, your markup. If the basic information is missing, you get skipped before a human ever sees you.

So for every product, ask yourself if an agent could answer these without emailing you:

• What is it, exactly? Materials, size, what it works with, what's in the box.
• How much is it? Price, currency, discounts, subscription terms. "Contact us for pricing" is basically a dead end.
• How does it compare? A simple comparison table against your main alternatives, and yes, include where you lose.
• When will it arrive? Shipping cost and delivery times for each region.
• What if it goes wrong? Return window, how fast refunds happen, who pays for return shipping.

Then make all of it easy for machines to read. This is the unglamorous, very tedious part, and it matters a lot.

1. Add structured data. Use schema.org Product and Offer markup for the name, brand, GTIN, price, availability and reviews. Add a MerchantReturnPolicy for your returns. Google lets you set one for the whole store and override it for specific products [4].
2. Fill in your feeds. Google Merchant Center is adding dozens of new attributes for conversational shopping, things like answers to common product questions, compatible accessories and substitutes [2]. If you're on Shopify, your catalog is what agents read in ChatGPT and Gemini [3].
3. Keep it crawlable. Put the facts in the page itself, not inside images, PDFs or tabs that only load after a click.
4. Keep one version of the truth. Your price on the page, in the feed and in the markup should match. If an agent finds three different prices, why would it trust any of them?

Tip: Ask an assistant to describe one of your products using only your product page. Whatever it gets wrong is what your page is missing.

Try it: Okara's GEO Agent works every day to get your brand cited in AI answers. Try the GEO Agent →

05 — Proof, not promises

Every brand says it's the best. Agents know that, so they look for two other things: who else says so, and what you've actually written down.

Reviews are back. For a while reviews felt like a nice-to-have. Not anymore. Start collecting verified reviews from your very first customers, don't wait until you "have scale." The best time to ask is right after the product proves itself, like the first delivery or the first week of use. Then show them on every product page as real text an agent can crawl, and get them onto the review platforms people in your category actually use.

Show, not just say. Publish the boring pages. Terms, return policy, warranty, support hours, and how to reach a real human. When two products look the same, the agent will pick the one whose policies are written down, because that's the safer bet for the person it's buying for.

Make everything verifiable. If you say you're sustainable, ethically made, clinically tested or SOC 2 compliant, link to the proof. A certificate, an audit, an approval. If there is nothing behind the claim, it's noise at best and a red flag at worst.

A quick way to check yourself: for every claim on your page, ask "where would an agent go to confirm this?" If the answer is nowhere, either add the proof or drop the claim.

Tip: Don't write your own reviews, don't reward only five-star ones, and don't hide the bad ones. Agents read the whole pattern. A few honest three-star reviews actually make your five-stars more believable.

06 — Get mentioned

What other people say about you counts more than what you say about yourself. An agent knows your homepage is marketing. A Reddit thread where five strangers argue about which one to buy is evidence.

You want to show up in three places.

1. Community threads. Reddit, Hacker News, niche forums. Real people answering "what should I buy for this?" is exactly what an agent is looking for when it builds a shortlist.
2. Expert comparisons. The "best X for Y" roundups, analyst lists, video reviews. Make reviewers' lives easy: send them your specs, a comparison sheet and a trial or a test unit.
3. Credible publications. Trade press, newsletters, podcasts with show notes. One real feature is worth more than a pile of paid placements.

A few rules so you don't get banned on the way:

• Answer the question first and mention your product second. And say you work there.
• Go where your buyers already ask. Search your category plus "recommendation" and read the top 20 threads before you post anything.
• If a thread gets your pricing or features wrong, correct it once, with the facts and a link. Don't get into a fight.

None of this is quick, and that's kind of why it works. Mentions are hard to fake, which is exactly why agents trust them.

Try it: Okara's Reddit Agent works the threads where your buyers ask for recommendations, every day. Try the Reddit Agent →

07 — Let the agents in

You can do everything in this handbook and still be invisible if your site is quietly telling agents to go away. Honestly, I'd check this one first, because it's the fastest fix.

The confusing part is that most AI companies run different bots for different jobs. One builds the search index, one fetches a page when a user asks a question, and one collects training data. Block the wrong one and you can drop out of AI answers without ever noticing.

Sources: OpenAI [5], Anthropic [7], Perplexity [8].

If you opt out of OAI-SearchBot, for example, you won't show up in ChatGPT search answers [5]. And the bots that fetch pages for a user, like ChatGPT-User and Perplexity-User, may not follow robots.txt at all, because a person asked for that page [5] [8].

If you want to be found, here is a robots.txt to start from. Training crawlers are their own business decision, so I've kept them separate.

# Let AI search and assistants read your site
User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

User-agent: PerplexityBot
Allow: /

# Training: your call (Allow or Disallow)
User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

Sitemap: https://yourdomain.com/sitemap.xml

Then go check your CDN and bot protection settings too. Your robots.txt can say yes while your firewall says no.

Can agents buy?

This part is still moving a lot. OpenAI launched checkout inside ChatGPT in September 2025, and by March 2026 it had moved checkout back to merchants' own sites and apps [6]. Meanwhile Google's checkout runs inside AI Mode and Gemini [2]. So I wouldn't bet everything on one protocol yet. Do the things that work no matter which one wins: guest checkout, no forced account creation, standard payment methods, and policies that are written down.

Lastly, decide your stance and write it down. Can agents browse your site? Sign up? Buy? Then make sure your fraud rules aren't quietly rejecting the orders you just said yes to.

Tip: Test again a day after you change robots.txt. OpenAI says search changes can take about 24 hours to show up [5].

Try it: Okara's free Robots.txt Generator and LLMs.txt Generator get this set up in a few minutes.

08 — Who owns the agent

Who owns the agent matters too, and so does what it's buying.

If it's something low stakes and low ticket, like milk or toilet roll, the agent will probably just decide on its own. And it learns your habits. If you keep buying the same milk from the same store, it will keep doing that until you tell it otherwise. So for these products the first order is everything. Win it, and you become the default.

If it's something expensive, like a laptop or a sofa, the agent does the homework and the person makes the call. It gathers the reviews, the terms and the comparisons, and comes back with a shortlist. Here you need two things: get on the shortlist, and then still win the human.

Then there's who built the agent. The ones from big platforms (think Meta, Amazon, OpenAI, Apple) each come with their own algorithms and their own incentives. A marketplace's agent might well favor the marketplace's own products. So find out which agents your customers actually use and what each one is optimizing for. Just like in the early days of SEO, whoever learns the rules first gets a real head start.

What stops working

Some of the marketing playbooks we rely on will stop working, or at least work a lot less.

Attribution. When an agent does the research, most of the clicks and touchpoints we used to track happen somewhere we can't see. Start measuring whether you get named and cited, not just who clicked.

Ads made only for humans. Ads won't disappear. They'll move to where the decision actually happens, inside the answer and right next to the buy button.

Gaming the system. Hidden text, fake specs, little instructions tucked into a listing to steer the agent. I get the temptation, but agents keep getting better at spotting this, and getting caught will cost you a lot more than the trick ever earned.

The cold start problem. Agents favor brands with long review histories, which is tough if you're new. So catch up fast! Get your first 50 verified reviews, get into a few comparison articles, and write clearer policies than the big players. Being the easiest brand to verify is something you can build in weeks, not years.

09 — Are you ready?

You don't need a six-month project to find out where you stand. An afternoon is enough. Here is my simple sequence:

1. Ask the assistant. Take the questions your customers actually ask and put them to ChatGPT, Claude, Gemini and Perplexity. How do they describe you? What do they get wrong? Who do they recommend instead?
2. Fix your product data. Complete specs, structured feeds and clear policies, everywhere you sell, not only on your own site.
3. Build a review engine. Ask every customer at the right moment, and publish reviews on the platforms your category uses.
4. Track your AI visibility. Measure how often you get mentioned and cited in AI answers, not just where you rank on Google.
5. Set your stance on agents. Decide if agents can browse, sign up and buy on your site, and then make sure your site actually lets them.

For step one, these are the prompts I'd start with. Swap in your own details:

> What's the best [category] for [use case] under [price]?
> [Your brand] vs [competitor]: which is better for [use case]?
> Is [your brand] legit? What do customers complain about?
> What is [your brand]'s return policy?
> Find me a [product] that ships to [city] by Friday.

For each answer, write down four things: were you named, was anything wrong, who got named instead, and which sources were cited. The wrong answers are your to-do list. The cited sources tell you where to go get your next mention.

So your job now is to impress humans and their agents both. People still decide which brands their agents trust. But more and more, the agents decide what actually gets bought.

Sixteen years ago, I spent weeks on forums before I trusted a shop enough to pay cash on delivery. Recently, I gave an AI assistant my card and let it buy groceries and a birthday gift. Somewhere in between, the trust moved from me to my agent. Make sure your brand is one it can check.

Marketing to AI Agents: Frequently Asked Questions

What is marketing to AI agents?

It's making your product easy for AI assistants to find, understand, trust and recommend when they're researching or buying for someone. In practice that means clear specs, structured data, real reviews, mentions on other sites, and a website agents are actually allowed to read.

Do I still need SEO if AI assistants do the shopping?

Yes, you still need it. AI assistants still crawl and index the web, so crawlable pages and clean markup matter as much as before. Think of GEO as building on top of SEO: it's about getting named and cited inside the answer, not just ranked.

How do I know if AI assistants recommend my product?

Ask them! Run the prompts from chapter 09 on ChatGPT, Claude, Gemini and Perplexity, and note whether you come up, what's wrong and which sources get cited. Do it every month so you can see if things are moving.

Should I block AI crawlers?

You can block the training crawlers if you want, that's a business call. But think twice before blocking the search and assistant crawlers. If you opt out of OAI-SearchBot, for example, you won't appear in ChatGPT search answers [5]. Chapter 07 has a robots.txt you can start from.

Can a new brand compete with established ones?

Yes, but not by waiting around. Agents like long review histories, so get verified reviews early, get into a few comparison articles, and write clearer policies than the incumbents. Be the easiest brand to verify.

Do I need to support agent checkout right now?

Not necessarily, not yet. The platforms are still figuring out how checkout should work, and OpenAI has already moved it back to merchants' own sites [6]. For now, make your own checkout easy for agents (guest checkout, standard payments) and turn on whatever AI channels your platform already supports.

Notes

1. Adobe, AI Traffic Drives 83% Higher Revenue Per Visit, October 29, 2026; figures cover US retail sites in September 2026, with AI traffic counted as clicks from AI sources.
2. Google, launch of the Universal Commerce Protocol and new agentic shopping tools, January 13, 2026.
3. Shopify, Millions of merchants can sell in AI chats, March 24, 2026.
4. Google Search Central, Merchant return policy (MerchantReturnPolicy) structured data.
5. OpenAI, Overview of OpenAI Crawlers.
6. Retail TouchPoints, AI Shopping Update: OpenAI Focuses on Discovery, Meta Launches Checkout, March 24, 2026.
7. Anthropic, help article on Anthropic's crawlers.
8. Perplexity, Perplexity Crawlers.