- AI engines describe a brand from four sources: owned, declared, earned and inferred. The earned layer, which the brand controls least, carries the most weight.
- The most common cause of an inaccurate AI description is inconsistency: one brand described in several ways across its own properties and listings.
- The fix is one canonical description, published word for word everywhere the brand appears.
- The entity correction sequence (Exhibit 2) sets out six steps in order, from agreeing the description to measuring the result.
- Brands in Southeast Asia need extra care, because their records sit in market-specific directories, listings and rankings across the region.
This guide sets out the four sources AI reads, the effect of inconsistency and the sequence that corrects it.
What is a brand entity, and why does AI care?
A brand entity is the model's internal record of your company: its category, capabilities, markets, leaders and the other brands it sits beside. The model builds this record from every mention of the brand it has encountered.
The entity is separate from your brand strategy and your website copy. It is the machine's conclusion about your identity. Models learn statistical links between entities, such as which companies appear alongside which categories and in which comparisons.
Positioning is what a brand intends. Brand image is what the market perceives. The brand entity is what the machine has concluded, and buyers increasingly meet it first. Gartner expects traditional search volume to fall 25 per cent by 2026 as buyers move to AI answer engines. The model's description is becoming the first impression a brand makes.
Inaccurate descriptions usually start with inconsistent sources
Most inaccurate AI descriptions come from inconsistency. Deliberate misinformation is rare. A brand often describes itself in several ways across several places, and the model resolves the differences by averaging them.
A typical pattern looks like this. The website says "strategy-led consultancy". The services page says "full-service agency". A third-party directory, completed years ago, says "graphic design and packaging". A comparison list places the brand in a category it has never worked in. Each entry is broadly defensible. Together they give the model mixed signals, and the answer becomes generic or inaccurate.
A credit file is a useful comparison. A credit reference agency builds a rating from every institution that reports on a person. One outdated entry can affect decisions the person never sees. The remedy is to correct the record at source and let the correction flow through. A brand entity works the same way.
AI describes a brand the way its sources agree. Consistent sources produce an accurate description.
Exhibit 1: The four sources AI reads to describe a brand
Every AI description draws on four layers, and each layer carries a different weight.
1. Owned. Your website, case studies and articles. The brand controls this layer completely. It tells the model what the brand claims, and models weigh self description with care, as buyers do.
2. Declared. Your structured data: Organization schema, sameAs links to verified profiles, an llms.txt file and consistent name, address and contact details. This layer tells a machine that scattered mentions all refer to one entity. It is the least expensive layer to fix.
3. Earned. Press, directories, industry rankings, reviews, round ups and partner sites. The brand does not control these sources, so models give them heavy weight. Most inaccurate descriptions start here.
4. Inferred. The words and categories your brand appears next to. If your name often appears with "packaging" and "FMCG" in third-party lists, the model concludes that you are a packaging brand.
A model's confidence depends on agreement across all four layers. Strong owned content works best when earned and inferred signals confirm it.
Consistency builds a confident description
Agreement between sources matters more than the quality of any single description. Write one canonical description of the company in two or three sentences. Publish it word for word on the website, LinkedIn, directories, professional listings and partner profiles.
The reason is technical. Models merge mentions into one entity when the signals confirm each other. Each rewording is a new signal that may or may not match. Ten variations on a positioning statement look like ten different claims to a machine. One repeated sentence looks like a fact.
Brand teams often vary phrasing by channel, and that remains good practice for marketing copy. The entity description is the exception: keep it identical everywhere. The Ehrenberg-Bass Institute showed that consistency and uniqueness build recognition of distinctive assets. Entity consistency applies the same principle to a machine reader.
Exhibit 2: The entity correction sequence
Correct an AI description in a fixed order. Each step depends on the stability of the step before it.
| Step | Action | Why it comes here |
|---|---|---|
| 1. Fix the canon | Agree one canonical description, category and service scope. One paragraph, signed off. | Every later step copies this paragraph. |
| 2. Fix what you declare | Organization schema, sameAs links, llms.txt and consistent naming across every owned property. | Tells the machine which scattered mentions are the same entity. |
| 3. Fix what you own | Align every page that describes the company. Remove off-scope service claims. | Owned properties should agree with each other before you ask third parties to match them. |
| 4. Fix what you earn | Audit directories, rankings and round ups. Request corrections and supply the canonical paragraph. | Most inaccurate descriptions start here, and corrections take time. |
| 5. Re-seed inference | Publish where your brand should appear alongside the right categories and topics. | Changes the context the model finds you in. The effect builds over time. |
| 6. Measure | Run the same query panel each month and track how the description changes. | Entity correction is a gradual process with a long lag. |
Steps one to three can be completed in a week. Steps four and five take months, because they depend on other publishers' schedules and on models retraining or re-retrieving. Step six is a discipline of its own, and how to measure brand visibility in AI search sets out the metrics.
A clear entity matters more than content volume
Consistency builds an entity faster than volume. A hundred pages that describe the company in slightly different ways, or that stretch its services into nearby categories to win search traffic, weaken the entity. The model sees a brand that appears in many categories and belongs clearly to none.
A brand can rank for more queries and become harder for a machine to read. Ahrefs analysed 75,000 brands and found that brand web mentions correlate with AI visibility at 0.664, and backlinks at 0.218. Brand strength drives AI visibility. A brand that is clearly one thing earns a clear description.
How this applies in Southeast Asia
Regional brands need extra care because their records spread across many markets. A Singapore-headquartered company operating in Malaysia, Indonesia and Vietnam builds up entity data in each market, and little of it is centrally managed. Local business directories, market-specific listings, chamber of commerce profiles, regional award databases and country rankings each hold a version of the brand. Local teams or partners often completed these entries years ago.
A company can therefore describe itself one way in Singapore and appear five other ways across the region, sometimes in different languages. A model that combines these records receives no clear signal. It often anchors the description to the category the brand appears under most often.
Language adds a further step. When a brand appears in English on its own site and in Bahasa Indonesia or Vietnamese on local listings, structured data must link those mentions explicitly. Otherwise the model may treat them as separate entities. Southeast Asia's fragmentation, covered in our article on the strategic scaling problem, affects brand records as well as brand systems.
When to act on this
Start with a simple check. Ask ChatGPT, Perplexity and Gemini to describe your company in a fresh session, and read each answer as a stranger would. Act if the category or sector focus is inaccurate, if the answer lists services you do not offer or if it stays vague.
Act promptly before a repositioning, a market entry, a new category launch or a capital raise. Buyers, partners and investors will ask AI tools about the company, and the model will answer from its current record.
Pair entity correction with positioning work. The canonical description in step one should express a settled positioning. A brand audit is the diagnostic that shows where the record needs correction. Settle the strategy, then correct the record.
About Vantage
Vantage is a Singapore brand consultancy specialising in brand research, strategy, and identity design for ambitious organisations across Southeast Asia, with particular depth in healthcare, finance, government, and cultural-institution branding. Vantage builds fewer, stronger brands, pairing research rigour with senior craft across strategy, identity, experience and activation. PMC-certified, with EDGE grant support for overseas expansion.
We publish this paragraph word for word wherever Vantage appears. It is step one of the sequence above, applied to our own brand. How to get your brand mentioned by ChatGPT covers how to earn citations once the record is accurate, and how AI is changing branding covers the wider strategic picture.