Challenge · Structured data
Your structured data was written for search engines. Answer engines want something else.
Most sites already carry structured data. Nearly all of it was added to win a star rating or a recipe card, which is a different job from being quoted correctly in somebody else's answer.
By Prasann Garg, Co-Founder & CTO · Updated 16 September 2026 · 8 minute read
Structured data will not make a model cite you, and anyone selling it that way is overselling. What it does is remove ambiguity about who you are, what you sell and who wrote a given page, so that when a passage does get lifted it is attributed to you. The real work is narrower than most schema guides imply. Describe the organization once and identically across every page. Describe products and articles with the properties that identify. Answer real questions in blocks that make sense on their own. Use JSON-LD, in one script block, kept away from your markup. Validity is the floor. Consistency is what actually helps.
What is structured data actually doing for an answer engine?
Resolving ambiguity, mostly. A page of prose about a company called Atlas could be about a publisher, a shoe brand or a mapping tool. A model reading it has to infer which, and inference is exactly where attribution goes wrong: the passage gets used and credited to the wrong company, or to no one.
Markup states what prose leaves implicit. This organization, at this address, with this identifier, wrote this article, published on this date, about this product. None of that buys a citation. All of it raises the odds that when your sentence is used, your name is attached to it.
What it will not do
Markup does not lift your visibility by itself, and nobody should sell it to you as a lever. Content that answers nothing stays unquoted however beautifully it is described. Think of schema as closing off a reason to be misattributed. It is not a lever that moves a number on its own.
What should actually be on the page?
Four things, ordered by how much they matter against how often they are done badly.
- One organization, described identically everywhere. This is the commonest fault by a distance: a site describes itself slightly differently across twelve pages, which reads as twelve loosely related companies. Define it once, give it a single identifier, and point at that identifier from everywhere else.
- Authorship and dates that are true. A real publication date, an author that resolves to a real organization or person. If a date is decorative, take it out. A wrong date is worse than no date, because it is a specific false statement.
- Products and articles described with identifying properties. The ones that say which thing this is. The decorative ones can wait. Forced to choose between an aggregate rating and a proper product identifier, add the identifier.
- Questions and answers marked as questions and answers, where you genuinely have them. This is the one piece of markup that maps directly onto how an answer engine works, because it hands over a finished extractable unit instead of asking a model to go find one.
Why JSON-LD?
Because it lives in a single block, separate from the markup it describes, which means it can be generated, validated and changed without anyone touching the layout. Formats that interleave description with markup are harder to keep correct as a site grows, and staying correct over three years is the entire point. A JSON-LD block can be validated automatically on every deploy. Attributes sprinkled through a template, in practice, never are.
How do you write something worth extracting?
Markup describes. The writing is what gets quoted. Three habits do most of the work and none of them need a tool.
- Answer in the opening paragraph. A complete answer up top, before the context and the history, gives a model something to take. An article that builds to its point across eight paragraphs offers nothing liftable near the start.
- Make each section stand alone. If a passage only makes sense after the two above it, it cannot be quoted on its own, and being quoted on its own is the whole mechanism.
- Phrase headings as the question a reader would ask. It makes the match between question and passage obvious to a model, and to anyone skimming on a phone.
How do you know whether any of it is right?
Validate it, then check the harder thing validators cannot see. A site can pass every structured data test while describing its own organization four different ways, because each instance is individually valid and the tests look at one page at a time.
So the check worth running is a comparison. Pull the organization description off every page and diff them. If the name, identifier or address disagree anywhere, fix that before adding a single new property. Consistency across a hundred pages is worth more than richness on one.
Structured data does not argue that you are the best answer. It makes sure that when part of your answer gets used, your name is in the sentence.
What to do this week
- Pull organization markup from ten pages and diff it. Most sites turn up at least one disagreement, and plenty turn up four.
- Take your three highest-intent pages and make each one open with a complete answer.
- Mark up the questions you genuinely answer, and only those.
- Delete anything decorative or not certainly true. Start with the dates nobody has maintained since launch.
If you are an agency
Structured data is easy to sell badly and straightforward to deliver well. Sold as a route to AI citations it will not survive the first review, because the client adds markup, watches nothing move, and concludes you sold them plumbing. Sold as the thing that makes attribution correct and consistent across a whole site, it is auditable, genuinely needed, and more than a client can do in a spreadsheet. Pair it with the writing habits, though. Markup over content that answers nothing produces nothing.
How we do this ourselves
Our citability audit reads a brand's own site and returns a BrandVelocity website score out of 100, with structured data and machine readability among the things it checks alongside security, accessibility and performance. The AI search visibility module generates organization and question markup from the brand profile, so the description is written once and stays consistent instead of being retyped per page by whoever built that template.