Decision Engine Optimisation (DEO)

What Is Decision Engine Optimisation?

Decision Engine Optimisation (DEO) is the practice of improving how AI systems evaluate your offer when helping a potential customer decide whether to buy it or choose it over competing offers.

People can use ChatGPT, Claude, and other large language models throughout a purchase journey. They can ask questions, discover products, research suppliers, and build a shortlist. They can also upload three offers and ask: “Which one should I choose?”

That last step deserves more attention from marketers.

A potential customer might already know your company. They might have spoken to your sales team and received a proposal. Your marketing has generated interest. Your sales process has produced an offer. Then the customer asks an LLM for its opinion before signing.

At that point, the AI acts as a decision engine. It can recommend your offer, recommend a competitor, or give the buyer reasons to reconsider the purchase altogether.

James Dooley coined the term in September 2026.

An AI recommendation can affect paid advertising

Jabez Reuben runs a performance marketing agency that advertises on Meta. His agency gets paid when it generates leads or sales.

He told me that he now checks what ChatGPT says about a potential client before deciding whether to work with them. If ChatGPT speaks negatively about the company or its product, he may decline the business.

His concern is what happens after the click.

An ad can reach the right audience, attract attention, and generate traffic at an acceptable cost. But a significant share of those potential customers will consult ChatGPT before converting. If the response discourages them from buying, the campaign’s economics can deteriorate.

An AI can also recommend the most expensive offer

James Dooley told me about a client he had recently won. When he asked why they had chosen him, the client explained that they had received three offers, uploaded them to Claude, and asked which one to buy.

James’s offer was the most expensive. The client chose it following Claude’s recommendation.

Claude played a role in selecting the supplier.

Together, these stories illustrate two forms of AI-assisted decision-making:

• Evaluating one option: “Should I buy this, or are there reasons not to?”
• Choosing between options: “Which of these offers should I accept?”

Decision engine optimisation needs to address both.

DEO addresses a distinct step in the sales funnel

Much of the discussion around generative engine optimisation (GEO) and answer engine optimisation (AEO) concerns whether a brand appears in AI-generated answers. Can the system find the company? Does it mention the product? Does it cite the website?

DEO focuses on the evaluation that leads to a purchase recommendation.

Consider the difference between these prompts:

Which agencies provide this service?

Here are three agency proposals. Given our requirements, which should we accept?

The second prompt starts with an existing shortlist. The buyer supplies the options. An agency could be included because of a personal referral, an event, an outbound sales conversation, or an existing relationship. The AI does not have to discover it.

This is why decision engine optimisation is its own discipline. It addresses a distinct task at a distinct stage of the sales funnel.

Whether someone classifies DEO as a subset of AEO, part of GEO, or another area of AI optimisation matters less than recognising the activity. Companies need to take responsibility for it.

Start with the offer document

The primary focus of DEO should be the actual offer sent to a potential customer.

That might be a proposal, quotation, pitch document, or statement of work. It is the material a buyer can upload and ask an AI to evaluate.

Start by examining how an LLM interprets that document. Does it understand what you are offering? Does it recognise the evidence supporting your claims? Does it identify uncertainty that you could resolve? Does it overlook something you consider an important advantage?

DEO can also extend to information outside the offer:

• Your homepage.
• Product and service pages.
• Pricing pages.
• Reviews on third-party websites.
• Your company’s reputation across other sources.

These become relevant when the AI uses them in its evaluation. If a buyer asks an assistant to research the suppliers as well as compare their proposals, the offer document is only part of the information being assessed.

The practical starting point remains the same: take the material your customer receives and examine what happens when an LLM evaluates it.

Build a benchmark for AI-assisted decisions

A company can begin with a straightforward benchmark.

Collect your own offer and, where available and appropriate to use, competing offers. Give them to several LLMs with a small set of realistic buyer prompts.

For example:

We are a medium-sized company looking for [service]. Our priorities are [criteria], and our constraints are [constraints]. Compare the attached offers and recommend which one we should choose. Explain the main trade-offs and identify any missing information.

Another prompt could ask:

For each of these three offers, explain the circumstances in which you would recommend it over the other two.

The second question is useful because the objective is not necessarily to win every possible comparison. You want to understand whether the system recognises the situations in which your offer is a strong fit.

Record which offer is selected, the reasons given, the objections raised, and any factual misunderstandings. Keep the prompts, documents, model details, and research settings so you can compare results over time.

A basic measure is the share of test runs in which your offer is the first recommendation. Track this alongside the recurring reasons for rejection.

Repeat the exercise regularly (monthly is a reasonable starting point) and after meaningful changes to your proposal.

To make the comparison useful, retain a consistent set of buyer scenarios. Test in fresh conversations and vary the order of the documents. Avoid declaring success based on a single favourable response.

When possible, compare the old and revised offers under the same current conditions. Otherwise, a change in results could reflect a change in the model rather than an improvement to your document.

This benchmark gives you a way to track performance in a defined set of AI-assisted evaluations. It does not, by itself, measure sales impact.

Make objection testing part of normal operations

You do not need competitors’ proposals to begin.

Upload your own offer and ask the LLM to evaluate it from the perspective of a realistic buyer:

Act as the CMO of a medium-sized company considering this offer. Our objectives are [objectives], and our constraints are [constraints]. Give me reasons why you would not choose it. Separate weaknesses in the offer from missing information and unclear wording.

Then ask follow-up questions:

Which concerns are supported by something in the document? Point to the relevant passages.

What additional evidence would help you evaluate the claims?

Which concerns would be most important for this buyer, and why?

Review the answers and decide which issues deserve action.

You might add evidence, clarify a deliverable, explain the pricing more precisely, or make a relevant qualification easier to find. You might discover that a valuable part of your service is barely mentioned in the proposal.

Sometimes the feedback will reveal a weakness in the underlying offer. A restrictive term, for example, might repeatedly count against you. That deserves commercial consideration, but it does not mean every AI objection should trigger a change to your product or contract.

Changing the substance of an offer belongs partly to broader offer optimisation, conversion optimisation, and competitive intelligence. DEO adds a specific perspective: how an AI system evaluates what the customer receives.

The feedback loop is a major part of the opportunity

One objection to DEO is that it sounds like good marketing. Make the offer persuasive, substantiate claims, and address objections.

The same argument could be made about paid search, organic search, or television advertising. They are all marketing, but each requires understanding a particular environment.

An LLM is a different evaluator from a human buyer. Test how it responds rather than assume that the presentation that convinces a person will produce the same evaluation from a model.

There is also a practical difference in the feedback loop.

Interviewing customers and prospects takes time. It requires recruitment, scheduling, thoughtful questions, and analysis. An LLM can return an evaluation of a proposal within a short interaction. You can repeat the exercise across multiple scenarios.

That makes it feasible to build a regular review process around AI feedback. Customer interviews remain valuable; the AI exercise supplies another source of information that is easier to run frequently.

Give DEO an owner

Companies should establish decision engine optimisation as an explicit responsibility. Whether they call it a project, a programme, or a service matters less than whether someone does the work.

That person needs access to current offers, an understanding of customer requirements, and the ability to turn findings into changes.

In many cases, this responsibility will lie in product marketing. It should not automatically be assigned to SEO.

An SEO team contributes useful expertise when the evaluation involves information found online. But the central task can involve proposal writing, positioning, commercial evidence, and purchase objections. The right owner is whoever can connect those elements and improve the material customers actually evaluate.

Understand what the tests can and cannot tell you

DEO is an emerging practice. A useful approach needs to account for several limitations.

Results can vary across models, prompts, conversations, and the information available to the system. A recommendation for one buyer profile does not establish that an offer will be recommended to everyone.

External reputation matters to a particular evaluation only to the extent that relevant information reaches the system. Uploading a proposal does not guarantee that the assistant will research the company online.

The explanations an LLM gives are useful feedback, but they are not a definitive account of the internal process that produced its choice. Treat them as hypotheses to investigate and test.

A model prompted to find objections can also produce weak or irrelevant objections. Human judgement is necessary before turning those responses into changes.

The first step is simple: take the offer you currently send to customers, give it to an LLM with a realistic buyer brief, and ask whether it would recommend accepting it.

Your potential customer can already do exactly that.