The agent has offered to add schema to every page. Before approving the change, open a page and read what it says. The markup will need to describe those facts, and some of the facts a buyer needs may still be missing.
What the pages say determines the order of work. Before approving a site-wide change, establish what markup does, whether Google's AI features require it, and how someone will check the agent's output. The same review needs to happen when the page changes later.
Schema markup labels facts your page already shows
Google asks you to describe the content a visitor can read on the page. Its structured data guidelines require the markup to represent that content truthfully.
Think of the JSON-LD and the visible text as two descriptions of the same page. One uses terms from a defined vocabulary; the other explains the company to a person. Adding a field to the first description does not establish a fact missing from the second. The page has to carry the information that the field represents.
This matters when an agent fills a template. A field can look appropriate for a business without being supported by the particular page. Review the value against what a visitor can find there. If the page lacks the fact, the next task is to establish and write it before adding its label.
Google also says that structured data does not guarantee an appearance in search results. A successful markup check shows whether the code describes the page correctly. It does not establish that Google will display a rich result for it.
For a business with a physical location, the local business schema guide covers the corresponding page details. Use that narrower guide when the location is the subject of the page. The principle remains the same: the markup describes the facts the visitor sees.
AI answers ask for no special schema, only indexed pages
Google's requirements for AI Overviews and AI Mode begin with the page's search eligibility. A page used as a supporting link must be indexed and eligible to appear with a snippet. Those are the checks to make before commissioning special markup for either feature.
Its AI features guidance also says those features require no special schema.org markup or additional optimization. That gives a founder a useful way to assess a proposed AI schema project. First establish whether the relevant pages can appear in Google at all.
Adding markup cannot substitute for that eligibility check. A team can finish the JSON-LD change while the page it describes remains outside the index. The page would still be missing from search.
Google does ask for the structured data to match the visible text. Read existing markup alongside the page to find differences. Correcting a stale description is useful work; Google's advice gives no reason to create a separate account of the business just for AI Overviews.
Keep the scope of the guidance clear when reviewing the work. These requirements concern Google's named AI features. They explain why a special markup package is not the starting point for those features; they do not promise that an eligible page will be selected in an answer.
Write the facts buyers check before you mark them up
A page needs enough information for an assistant to explain why the company belongs in its answer. Access lets the assistant read; the facts give it something to use.
An August 2026 review followed an assistant onto a startup's site. The assistant read the information and still left the company out. At that point the problem was the case the page made for the company. Labelling the same text again would leave that case unchanged.
The route to an answer helps locate that distinction. A company can be recalled from memory, included in a search query, found in results, read, shortlisted and picked. Markup concerns the point where the page is read. It cannot do the work of every earlier and later stage.
Before adding it, read the page as a buyer deciding whether the company can help. Does the text explain the work and support its claims? A number needs a source, and an example needs enough context to show what happened. That information belongs in the page itself.
The review should distinguish a finding from an achieved outcome. Discovering a missing page, publishing a revision and gaining a customer are different facts. Keep the markup specific to the finding, publication or result the page describes.
A coding agent should mark up only what each page shows
Give the agent the published page as its source, then review the proposed markup against that page. The instruction should specify both what it may describe and how the change will be checked.
Start with the marketing-site repository. The change can arrive as a pull request, where the owner can compare each proposed value with the page before merging it. This keeps the review attached to the actual code that would be published.
An instruction alone does not establish that every generated field is correct. Experience with coding agents shows why someone still reads the changes: agents can follow the intended rule on most pages and miss it on another. A site-wide edit deserves a page-by-page comparison.
- Constrain the source. Ask the agent to use only facts visible on each page, so generated fields have a text to check against.
- Inspect each value. Compare the proposed JSON-LD with that page, keeping unsupported details out of the change.
- Validate the result. Run Google's validation and read the pull request before the markup becomes part of the live site.
Validation and editorial review answer related questions. The validation check examines the markup; the person reviewing it checks whether its description is true. Both belong before the merge, while the changes can still be corrected together.
Update the markup whenever the page copy changes
When a writer changes the product description, the old wording may remain in JSON-LD. Open both versions during the review. A corrected name in the visible page is still inconsistent if the markup continues to describe the previous name.
Treat the structured data as part of a copy change. When someone reviews the revised page, they should also compare the fields that describe it. Google's requirement for a true representation continues to apply after the first implementation.
That is why the local business guide includes maintaining the visible content and its structured data together. A correct first version is only the starting record. Later edits can create differences even if nobody touches the markup deliberately.
During an AI search review, read the schema alongside the site's markdown copies and llms.txt. Each presents information from the site in a different form. Compare them with the current pages, looking for a name or claim that changed in one place and stayed behind elsewhere.
After the pages change, ask the buyer questions again and read the resulting answers. That check addresses whether assistants use the revised information. It gives the team a separate observation from the fact that the markup now matches the page.
Markup earns most where Google still shows a rich result for it
Choose markup by the search feature a page can qualify for and the facts it can describe accurately. Google's supported result types have changed, so an old checklist may assign work to a feature that is no longer available.
Google changed two familiar result types in its 2023 announcement. FAQ rich results became restricted to well-known, authoritative government and health websites, and How-to rich results were deprecated. Check those limits before allocating engineering time to either feature on a B2B startup's site.
A useful FAQ can still answer a buyer's question on the page. That writing decision is separate from whether Google shows a FAQ rich result. The FAQ SEO guide addresses the section itself, while this guide concerns how its facts are labelled.
A physical business has a more specific reason to examine local business markup. A young B2B site with few indexed pages has a different first job: make its important pages available and give them useful evidence. The same implementation task can have a different priority on those two sites.
Decide what the markup is meant to support before assigning it across the whole site. A relevant result type, a page that can appear in search and accurate visible information give the work a concrete purpose. Completing a schema checklist alone does not establish those conditions.
Example: checking what an AI-ready site gave markup to describe
One client review found a completed AI-readiness checklist alongside very little indexed content or evidence in the machine-readable copy.
| Check | Finding |
|---|---|
| AI-ready checklist | Done well |
| Pages indexed by Google | Two |
| Customer names in the markdown copy | Zero |
| Logos or testimonials in that copy | Zero |
| Numbers in that copy | Zero |
The findings put indexing and page evidence ahead of more labels. Once those pages contain supported facts, markup has a fuller account to describe.
Check the page before the markup
Inspect search eligibility
Open URL Inspection for the page and establish its index status before planning markup around a possible search appearance.
Write the missing facts
Add the information a buyer needs to judge the company, retaining the source for every statistic used.
Label visible information
Ask for JSON-LD that describes the published text, then validate the proposed markup with Google's check.
Review both versions together
Compare the pull request with the page before merging, and include that comparison in later copy changes.