How Do You Sell to an AI Agent? A Guide for Businesses

AI agents are beginning to research, compare and buy on people's behalf. When the first buying decision isn't made by a person, the goal shifts from persuading machines to being easy for them to evaluate. Here is how to prepare.

For decades, marketing has been built around persuading people. Write the right headline. Show the right image.

Tell the right story. Build trust. Create desire.

Remove objections. Ask for the sale. But what happens when the person making the initial buying decision isn't a person?

Imagine a customer says to their AI: "Find me the best accounting software for my small UK business. I have six employees, use Stripe and need payroll. Keep it under £100 a month." The AI researches the market. It creates a list of possible providers.

Checks features. Reads reviews. Compares pricing.

Looks at integrations. Eliminates unsuitable products. Then returns with one or two recommendations.

Perhaps, eventually, it can even open the account. The customer may never see most of the marketing created by the companies that were considered. So how do you sell to an AI agent?

The answer probably isn't to become better at persuading machines. It's to become much easier for machines to evaluate.

AI changes the buyer journey

A traditional online buying journey might look like: awareness, then research, then website, then consideration, then conversion. Businesses have built entire marketing strategies around influencing each stage. An agent can compress that journey.

The human provides the objective. The agent performs much of the research. The agent filters the options.

The human approves the recommendation. The agent may eventually perform the transaction. The journey becomes: intent, then agent, then decision, then approval, then action.

That moves a significant amount of influence away from the website and into the agent.

Your customer may have a buying agent

Businesses already employ people to buy on their behalf. Procurement teams do it. Personal assistants do it.

Travel agents do it. Financial advisers do it. AI potentially gives everyone a version of that capability.

Imagine having a highly patient purchasing assistant that can read 500 product pages, compare 100 prices, analyse thousands of reviews, check every specification, remember your preferences, calculate the real cost, identify hidden fees, and do it in seconds. That could be extraordinarily useful for consumers. It could also be deeply disruptive for businesses whose marketing depends on capturing attention.

Because an AI doesn't necessarily shop like a human.

AI doesn't need to be impressed by your homepage

A beautiful website still matters when a human visits. But an agent may care very little about the hero image you've spent three weeks choosing. It may not care that your button is green rather than blue.

It may not respond to urgency in the same way a person does. "ONLY THREE LEFT!" "SALE ENDS MIDNIGHT!"

"MOST POPULAR!" An agent can potentially examine the underlying evidence. Are there really only three left?

Is this genuinely the best price? Is the "most popular" product actually appropriate for its user? The techniques used to influence human attention may become less effective when an AI is doing the filtering.

That means businesses may need to compete on something much less comfortable: the substance of the offer.

The first challenge is getting into the consideration set

An AI can't choose a business it doesn't know exists. That makes visibility the first problem. When a user asks an AI for a recommendation, the system needs enough information to identify potential providers.

This is where SEO, GEO, brand authority and entity recognition start to converge. An AI needs to understand that your company exists, what you sell, who it is for, where you operate, what makes the product relevant, how much it costs, and whether the information is trustworthy. If your business is absent or poorly represented in the information environment the AI uses, you may never enter the comparison.

For businesses, that means AI visibility increasingly becomes a commercial issue rather than simply a marketing experiment.

Then the agent needs to understand your offer

Getting mentioned isn't enough. The agent needs enough information to evaluate you. Suppose someone asks: "Find me a project management platform for 25 people for less than £200 a month that integrates with Slack and allows external client access."

If your website clearly publishes pricing, user limits, integration details, permissions, client access, contract terms and feature information, the agent has something to work with. If your site says: "Contact sales to discover a solution tailored to your unique business journey," the agent has a problem. It doesn't know the price.

It doesn't know whether the requirement is met. It doesn't know whether contacting sales is worth its user's time. Another provider may simply be easier to evaluate.

In an agentic world, information friction could become conversion friction.

Hidden pricing may become more expensive

Businesses have legitimate reasons for not publishing prices. Complex services aren't always easy to package. Enterprise contracts vary.

Projects need scoping. But hidden information creates uncertainty. Humans sometimes tolerate uncertainty because they are emotionally interested in a product.

Agents may be more ruthless. If the user's requirement is "under £100 a month" and the agent cannot establish whether your product meets that requirement, it may remove you from consideration. This doesn't mean every company must publish a single fixed price.

It means businesses may need to provide enough information for agents to understand whether an offer is broadly suitable. For example: "Plans start at £79 per month." "Typical projects range from £2,000 to £5,000." "Enterprise pricing depends on user volume." "Free for teams under five." These statements reduce uncertainty.

Your claims need evidence

Humans are used to marketing language. "Industry leading." "Best in class." "Revolutionary." "Trusted." "Premium." An AI agent has little reason to accept those claims at face value.

If it is evaluating products for a user, it can potentially look for evidence. That could include independent reviews, customer ratings, case studies, certifications, credible media coverage, third-party comparisons, technical documentation, published research, customer outcomes and other external references. That makes verifiable reputation extremely important.

The businesses most visible to AI may not simply be those making the loudest claims. They may be those for which the clearest body of evidence exists.

Reviews become data

A person might read five reviews before buying. An AI can potentially analyse thousands. That changes what reviews represent.

They aren't only social proof displayed beside a product. They are a dataset about customer experience. Imagine an agent evaluating two hotels.

Hotel A has 4.8 stars. Hotel B has 4.6. A simple ranking might favour Hotel A.

But the agent knows its user is travelling with two young children. When it analyses the reviews, it discovers Hotel B is repeatedly praised for family rooms, helpful staff and quiet evenings. Hotel B may therefore be the better recommendation for that particular person.

This is why the agentic internet could be far more personalised than search rankings. There isn't necessarily one "best" product. There is the product that best satisfies this user's constraints.

Specifications matter more when machines compare them

Product and service data is often treated as administrative information. In an agentic world, it becomes marketing infrastructure. An agent may compare dimensions, materials, ingredients, compatibility, delivery times, availability, energy efficiency, warranties, service levels, contract lengths, cancellation terms, locations, qualifications or hundreds of other attributes.

If those attributes are difficult to access or inconsistent across sources, the agent has less confidence. Accurate, structured product information therefore becomes commercially valuable.

Stop thinking only about pages

Most digital marketing still revolves around pages. Homepage. Service page.

Product page. Landing page. Blog post.

But an AI agent doesn't necessarily experience a business as a collection of pages. It may experience it as information and capabilities. Information answers what this is, how much it costs, whether it is suitable, whether it is available and whether it can be trusted.

Capabilities answer whether the agent can check availability, get a quote, book it, buy it, cancel it, return it or track it. That distinction matters. The businesses best prepared for agents may not simply have excellent websites.

They may have excellent information architecture and action architecture.

Make actions machine-accessible, but controlled

There is an obvious temptation here. If agents are going to buy things, businesses should let them do everything. That would be a mistake.

Agent accessibility needs to sit alongside security. Businesses will need to decide which agents can interact, how agents identify themselves, which information is public, which actions require authentication, what users must approve, which transactions have limits, what gets logged, and when human intervention is required. The goal isn't unrestricted agent access.

It's controlled agent access. An agent should be able to perform useful actions without gaining unnecessary authority. This is the same least-privilege principle businesses should already be applying to their own AI agents.

Trust may become a machine-readable advantage

Humans use countless subtle cues to decide whether they trust a business. Design. Language.

Brand familiarity. Photography. Recommendations from friends.

Agents may use a different combination of signals. They might examine domain history, company information, reviews, independent references, returns policies, contact details, security practices, professional accreditations, consistent business data, customer complaints, terms and conditions, and previous interactions. We don't yet know exactly how future purchasing agents will weigh these signals.

Different agents will almost certainly behave differently. But the direction is clear. Trust needs evidence.

Brand still matters

It would be easy to conclude that agents make brands irrelevant. That's unlikely. Brands are information too.

A strong brand creates recognition, reputation, reviews, mentions, citations, customer preference, direct demand and evidence of experience. And the human can always instruct the agent: "Only show me brands I know." "Don't buy from X." "I prefer Apple." "Use the company we bought from last time." "Find me something similar to Patagonia." Human preferences don't disappear because an agent is involved.

The agent simply applies them. Strong brands may therefore remain extremely powerful. But weak businesses may find it harder to compensate for poor products with clever advertising if an agent can analyse the underlying evidence at scale.

Advertising won't disappear, but it may change

Where there are commercial decisions, businesses will attempt to influence them. AI agents won't change that. We are likely to see new forms of paid visibility, commercial integrations, preferred providers and sponsored recommendations.

The important question will be transparency. If an agent recommends Product A because it genuinely believes Product A best satisfies my request, that is one thing. If Product A paid the agent provider for placement, that is another.

Consumers will need to know the difference. AI companies that become trusted purchasing intermediaries will have to manage that tension carefully. Their commercial interests may not always align perfectly with their users' interests.

The agent could negotiate

Agentic commerce gets even more interesting when agents move beyond selecting products. Imagine telling your AI: "My broadband contract expires next month. Find a better deal. If my current provider will match it, stay. Otherwise switch me." The agent could review your current contract, research competitors, compare offers, contact your provider, request a retention offer, evaluate it, ask you for approval, and complete the switch.

Now imagine millions of consumers doing this. Entire industries currently benefit from customer inertia. Insurance.

Broadband. Energy. Subscriptions.

Banking. Software. Agents could dramatically reduce that inertia.

That may force businesses to compete much more continuously on actual value.

Loyalty could become harder to fake

Humans forget. Agents don't have to. An agent could remember what you paid last year, when your introductory offer ends, whether the price increased, what competitors charge, whether you had a bad support experience, and whether the company promised something it failed to deliver.

That gives the consumer potentially enormous negotiating power. Retention strategies based on customers not noticing may become less effective. Businesses that create genuine long-term value could benefit.

B2B buying may change even faster

The implications aren't limited to consumer shopping. Consider a company looking for a CRM, an accountant, cybersecurity software, a recruitment agency, cloud hosting, insurance, office equipment, marketing software or a supplier. AI could perform much of the early procurement process.

It could create requirements. Research vendors. Build comparison tables.

Analyse contracts. Read documentation. Check reviews.

Compare prices. Identify risks. Then present a shortlist to the decision-maker.

That means B2B businesses need to think about whether AI can accurately evaluate them too. The PDF brochure hidden behind a lead form may become less useful than transparent, structured, verifiable information.

How do you actually optimise for an AI buyer?

There is no universal formula. Anyone claiming to know exactly how every future purchasing agent will rank every business is getting ahead of the evidence. But there are sensible foundations businesses can work on now.

1. Be easy to discover. Make sure major search and AI systems can establish that your business exists and understand what it does.
2. Be easy to understand. Use clear language about products, services, audiences and locations.
3. Be easy to compare. Provide useful specifications, features, pricing information and terms.
4. Be easy to verify. Support claims with evidence, reviews, case studies, credentials and independent references.
5. Be technically clear. Use appropriate structured data and maintain consistent information.
6. Be current. Old prices, discontinued products and outdated policies create uncertainty.
7. Build reputation outside your own website. An agent evaluating your claims may look elsewhere for confirmation.
8. Think about agent-accessible actions. Could a legitimate agent check availability, request a quote, book or purchase?
9. Protect those actions. Agent access needs authentication, permissions, limits, approvals and logging.
10. Monitor AI visibility. Find out what AI platforms currently say about you, your products and your competitors.

The aim isn't to manipulate AI

This is important. There will inevitably be businesses promising techniques to "hack" AI recommendations. That isn't where we think the durable opportunity lies.

The better question is: how can we make our business the easiest credible option for an AI to evaluate accurately? If your product is good, make the evidence accessible. If your pricing is competitive, make it understandable.

If customers love you, make sure that reputation exists in places machines can discover. If you have specialist expertise, demonstrate it. If your product solves a particular problem exceptionally well, describe that problem clearly.

Optimising for AI should ultimately mean reducing ambiguity between what your business genuinely offers and what machines can understand about it.

Marketing may become less about capturing attention

Digital marketing has spent decades fighting for attention. Search ads. Display ads.

Pop-ups. Retargeting. Social feeds.

Email subject lines. Notifications. Every company trying to earn a few seconds of human focus.

Agents introduce the possibility of something different. A machine isn't necessarily short of attention. It can analyse 100 options.

The scarce resource becomes relevance. Does your product satisfy the requirement? Can the agent establish that confidently?

Can it verify your claims? Can it transact safely? That could shift some areas of marketing away from "How do we make them look?" towards "How do we make ourselves the most appropriate answer?"

That would be a profound change.

Your next customer might send an agent

AI agents are still early. Most purchases today are still made by people navigating websites themselves. Businesses shouldn't redesign their entire commercial strategy around a future that hasn't arrived.

But they shouldn't ignore the direction either. AI systems are already influencing what people discover. They are already recommending products and companies.

Agents are beginning to move from recommendation into action. The next step is obvious. People will increasingly delegate parts of buying to AI.

When that happens, businesses will have a new kind of customer journey to understand. Not just: how do we persuade this person to buy from us? But: can their AI find us, understand us, verify us, compare us and transact with us?

And ultimately: does the evidence give it a good reason to choose us? That may become one of the most important marketing questions of the agentic internet.

Where to go next

• What happens to SEO when AI agents browse the web for us? The discovery half: can an agent find and understand you in the first place?
• How to get your business recommended by ChatGPT Practical GEO groundwork for being surfaced and cited by AI systems.
• Your AI strategy and your cybersecurity strategy are becoming the same thing If agents can transact with you, agent access needs authentication, limits and approvals.

Related: What happens to SEO when AI agents browse the web for us?.