Before I worked in AI, I studied branding. One idea from that training has come back this year with more force than I expected: the difference between brand identity and brand image.
For most of my career, the gap between the two was a marketing problem. It is now becoming a data problem. Your brand identity is now a dataset, and whoever holds the cleanest version of it decides how you are described.
What is the difference between brand identity and brand image?
Brand identity is what an organisation actually is: what it does, how it does it, what it stands for, what it charges, who it is for. It belongs to the sender. Jean-Noël Kapferer’s brand identity prism, one of the frameworks most branding students meet early, is built around exactly that split between the brand as sent and the brand as received.
Brand image is the received side: what people hold in their heads. Kevin Lane Keller’s 1993 paper on customer-based brand equity defines brand image as the set of associations a customer holds about a brand. It is formed from everything they encounter, most of which the brand does not control.
The work of branding, for decades, was managing the distance between the two. You could not dictate the image, but you could shape it: advertising, packaging, service, reputation. The image was formed in people’s heads, and people could be reached.
What has AI changed?
A growing share of first impressions no longer forms in someone’s head. It is assembled for them.
When a customer asks an AI assistant which accountant, hotel or builder to use, the assistant writes the first description they read. It does that by pulling together whatever sources it can find and trust. That description is brand image, produced by a machine, delivered before the customer has seen anything the business made.
When we checked what assistants draw on for ordinary Caribbean queries earlier this month, 23 sources came back across three everyday categories, and exactly one was a business’s own website. The businesses were being recommended by name. They simply were not the ones supplying the words.
That is the shift. Image used to be formed from impressions. It is increasingly formed from data, and mostly from data someone else published.
Why is structure now the bottleneck?
If an assistant is going to describe you accurately, the true version of you has to exist somewhere it can read: specific, consistent, current, and checkable against other sources. A beautiful website that says “quality service since 1998” gives it almost nothing to work with. A plain page that states what you do, for whom, where, at what price and with what exceptions gives it a great deal.
The same constraint shows up inside businesses trying to put AI to work. Sanity, the content platform, published research in May 2026 based on its own platform data. It found that 64% of teams that had cleared every other hurdle still could not automate, because their content was not structured cleanly enough for AI to act on. Their summary is blunt: the spending goes to AI tools, but the bottleneck is how the content is organised. It is one company’s data about its own customers, but it matches what we see.
Whether the reader is a customer’s assistant or your own team’s, the requirement is the same. The identity has to be explicit before anything can be built on it.
Where does a business’s identity actually live?
Not in the brand guidelines. In most organisations I have worked with, the real identity is scattered across people and places:
- In the heads of the two or three people who have been there longest.
- In the answers the front desk gives to the same ten questions every week.
- In the way a senior person writes a quote, and the exceptions they make without writing them down.
- In WhatsApp threads, spreadsheets, email chains and a system nobody wants to touch.
- In reviews, written by customers, describing what the business is actually like.
Almost none of it is in a form a machine can read, and much of it is not in a form a new employee can read either. That was always a problem. It is now also a visibility problem.
How does service design get that identity out?
This is where the second half of my background comes in. Service design has a set of methods built for exactly this job: understanding how an organisation really works, as opposed to how it describes itself.
The best known is the service blueprint, described by G. Lynn Shostack in the Harvard Business Review in 1984. It maps what the customer sees, what happens behind the scenes to deliver it, and the handovers between them. Alongside it sit simpler habits: sitting beside the people doing the work, following a request from first message to invoice, and noting where information gets retyped, lost or held in one person’s memory.
For forty years those methods served one purpose: helping an organisation understand and improve its own offering. That purpose has not gone away. It has gained a second audience. The same work that shows a business where its service breaks down also produces the most accurate, specific description of what the business is. It is exactly the material an assistant needs, and exactly what generic marketing copy lacks.
In practice, the output is a set of plain question-and-answer units: one question, one checked answer, evidence where there is some, an owner, and a date for review. We call them knowledge objects. Others call them answer cards. The name matters less than the discipline.
What does one knowledge base, inside and out, look like?
Once the identity is captured properly, it can serve both sides from one source:
- Inside, it becomes the reference your own AI assistants and your staff work from: the price list, the policies, the way things are done, the same answer every time.
- Outside, an approved part of it becomes your website’s answers, your Google Business Profile, your listings and your structured data: the version of you that assistants read.
The word that matters is approved. Not everything inside should go outside. Credit terms, discount rules, customer names and internal notes stay in. The Sanity research found that 11 of the 12 enterprise teams it interviewed had no written AI use policy, and its suggested fix is a single page answering three questions: what can be shared with AI, who reviews what it produces, and what stays human-only. That page is also, in effect, the line between your brand identity and the part of it you choose to make public.
This is also why I do not believe the website is dying, whatever the headlines say. Its role is changing. It stops being the whole brand and becomes the public, verifiable edition of a knowledge base that lives underneath: the place where the facts are stated plainly enough for a person or a machine to check.
What should a business do first?
- Measure the image you have. Ask the assistants your customers use the questions your customers ask, and write down what they say and which sources they cite. Date it.
- Collect the identity you have. List the questions your team answers every week, and who answers them. That list is the start of your knowledge base.
- Compare the two. Where the assistants are wrong, vague or quoting a directory instead of you, you have found the gap between identity and image.
- Write the one-page rules before anything is published or automated.
- Publish the approved answers where they will be read, then measure again on a fixed date.
None of this needs new technology to start. It needs the discipline branding people have always argued for: know what you are, say it consistently, and make sure the version of you that reaches people is the true one. The difference now is that one of the people you need to reach is a machine.
Frequently asked questions
What is the difference between brand identity and brand image?
Brand identity is what an organisation actually is and intends to be, defined from the inside. Brand image is how it is perceived from the outside, formed from everything people encounter about it. Branding has always worked on the gap between the two. AI assistants now form a growing part of that image by assembling descriptions from whatever sources they can find.
Why call brand identity a dataset?
Because AI assistants can only describe a business from what they can read. If the true facts about a business are not written down clearly, consistently and somewhere checkable, the assistant fills the gap from other sources. A structured record of what the business does, for whom, where and at what price is, in effect, its identity in a form machines can use.
What does service design have to do with AI visibility?
Service design methods such as shadowing staff and service blueprinting were built to reveal how an organisation really works. That same work surfaces the specific, first-hand knowledge that makes a business distinguishable, which is exactly what AI assistants need and what generic marketing copy lacks.
Does this mean my website no longer matters?
No. Its role changes. It becomes the public, verifiable edition of your knowledge base: the place where your facts are stated plainly enough for a customer or an assistant to check. First-party websites can still be a trusted source when they are built for that job.
Where should a small business start?
Write down the ten questions your team answers most often and the answers you actually give. Then ask an AI assistant those same questions about your business and compare. The differences show where your image has drifted from your identity, and what to publish first.
Sources
- Kevin Lane Keller, Conceptualizing, Measuring, and Managing Customer-Based Brand Equity, Journal of Marketing 57(1), 1993, pp. 1–22.
- Jean-Noël Kapferer, the brand identity prism, set out in his work on strategic brand management (first published in the 1980s; see The New Strategic Brand Management, Kogan Page).
- G. Lynn Shostack, Designing Services That Deliver, Harvard Business Review, January 1984.
- Sanity, AI Content Operations Report, May 2026. Based on Sanity’s own platform data (1.46 million AI tool calls, September 2025 to April 2026) and 12 interviews with content leaders; describes early adopters on one platform.
- UX Caribbean, The Button Didn’t Die. It Moved., 11 September 2026. Source counts from US-geolocated web search results for everyday Caribbean queries; method and limitations are stated in that post.