On May 7, 2026, Google retired FAQ rich results, and with them the last good reason to treat schema as a SERP-display trick. What survived the change is the part that always mattered: structured data as the machine-readable identity AI engines check before they cite a brand. We audit, architect and implement that identity — Organization, Service, knowsAbout and the entity wiring underneath — so ChatGPT, Claude, Gemini, Perplexity and Google's AI answers can name you with confidence, in English and Spanish, at a fixed price you can see below.
Two events this year redefined what structured data is for. Google's March 2026 core update narrowed rich-result display for FAQ, Review and How-To markup that decorated non-primary content — the classic "schema for the dropdown" play. Then on May 7, Google deprecated FAQ rich results entirely, following How-To before them. If an agency is still selling you schema as a way to win stars and expandable boxes on the results page, they are selling last year's product.
Here is what the same period quietly confirmed. Sites with clean, accurate entity schema saw improved citation rates in Google's AI answers after the March update. Google's search team said in April 2025 that structured data gives an advantage, and Microsoft's Bing team confirmed in March 2025 that schema helps its LLMs understand content for Copilot. The display features died; the verification layer got more valuable. That is the layer we build.
And one honest nuance most pitches skip: for AI extraction, your visible text is what gets quoted. The JSON-LD does not get extracted — it tells engines who is speaking, what the page is, and whether the claims belong to a coherent, verifiable entity. Schema without content alignment is noise; content without entity schema is anonymous. The work is making both agree.
The flip side of the update matters just as much: legacy markup is no longer harmless. Schema that contradicts visible content, FAQ blocks bolted onto pages for a dropdown that no longer exists, Review markup on editorial comparisons — the March update demoted some of these patterns algorithmically and manually actioned others at scale. A surprising share of the sprints we scope begin with removal: stripping years of accumulated markup-for-display before building the verification layer in its place. If your site has been through three agencies and two CMS migrations, assume the cleanup phase is half the value.
None of this means rushing to delete markup either. FAQ and How-To types remain valid schema, Bing and the AI crawlers still parse them, and removal only pays where the markup misdescribes the page. The discipline is simpler than the panic suggests: keep what describes real content, remove what existed for a display feature that no longer does, and put the freed effort into the entity layer that the engines now actually read.
Entity problems rarely announce themselves. They show up as a quiet tax on every other marketing effort, and most teams meet them in one of these five forms:
Every one of these is fixable, and the fix runs through the same place: a single, coherent, machine-readable identity that every page reinforces. That is what the sprint builds. A useful self-test before you spend anything: ask two or three engines what your company is and what it does. If the answers disagree with each other, or with your current homepage, you have just watched the problem happen — and you now also have the before-snapshot the sprint will be measured against.
audit
Every template and key page inventoried: what schema exists, what validates, what contradicts the visible content, and what the March 2026 rules now treat as noise. You get a findings document with each issue tied to the page it lives on, ordered by impact, including the markup worth removing — an audit that only adds is not an audit.
identity
A single, authoritative Organization graph: legal and brand names, founding date, logo, sameAs links to the profiles that disambiguate you, and a knowsAbout set that declares the topics you genuinely cover. This is the identity every AI engine checks before it names you, built once and referenced site-wide.
templates
Service, Product, Article, BreadcrumbList and the types your content actually warrants, implemented at template level so every new page ships with correct markup instead of depending on someone remembering. Delivered as code for your stack, not a PDF of recommendations.
entityhome
One clear canonical URL per core entity — each service, each product line — with supporting pages linked back using descriptive anchors, so machines can tell which page is the authority on what. This quiet structural work is where many sites lose disambiguation without noticing.
validation
Everything validated against current tooling, plus a baseline check of how engines describe you before and after. We treat changed AI descriptions as the success metric, because that is the point of the work — not a green checkmark in a validator.
Since the March update, the single highest-impact addition for most organizations is a property few
sites use: knowsAbout. It declares, in machine-readable form, the subject areas your
company and your authors genuinely cover. AI systems weigh topical authority when choosing sources
for a query category, and an organization that declares expertise in, say, email deliverability and
B2B SaaS marketing is measurably more likely to be cited for queries in those domains than an
identical organization that declares nothing.
The catch, and the reason this is craft rather than copy-paste: it only works when it is true. Declaring twenty topics you barely touch reads as the same spam the March update punished. We build your knowsAbout set from your actual content footprint — the topics where your site demonstrates real depth — and we wire it to author entities where your team has named, verifiable expertise. Done honestly, it is the cheapest authority signal in the stack. Done greedily, it is worse than nothing.
A worked example of the difference. A deliverability SaaS we would scope declares knowsAbout for email deliverability, SPF and DKIM authentication, and inbox placement — three topics its blog and docs cover in depth — each tied to a named author with a real publication history. It does not declare "artificial intelligence" because it published two AI think-pieces last spring. The narrow, true declaration wins citations in its actual category; the broad, aspirational one would dilute the signal everywhere and convince the engines of nothing. Topical authority is earned in the content and declared in the schema, in that order.
This sprint fits B2B SaaS and service companies whose buyers research by asking AI engines — and whose category has lookalike names, ambiguous acronyms or crowded comparisons where a hedging model defaults to someone else. It is also the natural second step after an AI visibility audit reveals that engines describe you vaguely, confuse you with a competitor, or cite you without naming you. If your audit showed wrong pricing or retired features in AI answers, entity work is usually where the fix starts.
For bilingual brands the case is stronger still. Most companies serving English and Spanish markets
run two half-identities that contradict each other in small ways — different descriptions, mismatched
names, schema in one language only. We unify the entity across both, with correct
inLanguage values and hreflang coherence, so engines in either language resolve to the
same verified organization. The AC Group has earned attention online for 27 years and operates
natively in both languages; this is not a translation bolted on at the end.
Where it fits less well, said plainly: if your AI visibility problem is that nobody on the wider web mentions you, entity work alone will not fix it — that is an earned-media problem, and we will tell you so in the audit rather than sell you a sprint that cannot move your number. Entity schema makes you verifiable; it cannot make you talked about. The honest sequencing is the six-step method: this sprint is steps three and four, and it pays best when steps one and two are done and step five is planned.
01
Week one: full entity and markup inventory, the findings document, and the target architecture agreed with your team before anything is written.
02
Weeks two to three: the Organization layer, template schema and entity homes implemented in your stack, in pull requests your developers can review, or hands-on if you prefer we ship it.
03
Final week: validation across tools, the before/after description check, and a maintenance one-pager so the next page your team publishes keeps the architecture intact.
delivered as code for your stack · astro, next.js, wordpress and most modern CMSs
Two working styles, your choice. Hands-on: we ship the implementation directly, in branches your team reviews, which suits lean teams without spare engineering cycles. Or advisory: we deliver the exact code and your developers merge it, which suits organizations with change-control requirements. Either way the deliverable is identical and the validation phase is ours — we do not hand off architecture and hope. And because everything ships as code rather than a recommendations deck, the work survives your next redesign: templates carry the schema forward, and the maintenance one-pager tells whoever builds the next page exactly what to keep intact.
Entity engineering is infrastructure, so its wins show up in how everything else performs. The first change is usually descriptive: within weeks of the identity layer shipping, engines that hedged start describing you specifically — right category, current positioning, your own framing echoed back. Because retrieval-based engines read the live web, Perplexity typically moves first, with the slower engines following as their indexes refresh.
The second change is attributional. Content that was being used anonymously starts carrying your name, because the passages engines were already extracting now belong to a verifiable organization. And the third is defensive, the one nobody sees: the mix-ups stop. When a model can cleanly separate you from the similarly-named competitor, their bad press stops bleeding into your answers and your features stop appearing in theirs.
On timing, the honest ranges: descriptive improvements typically surface within two to six weeks, attribution gains over one to three months as indexes and models refresh, and the defensive effects are immediate but only visible when the next confusion-event fails to happen. Anyone promising next-day transformation is describing a different, easier problem than the one you actually have.
We measure all three against the baseline prompt set, before and after, per engine. What we do not promise is a citation guarantee — no honest vendor can, because citation also depends on what the wider web says about you. What the sprint guarantees is that when engines check who you are, the answer is unambiguous, current and yours. That is the precondition everything else in a GEO program builds on.
entity & schema sprint
€2,900
One language. Full audit, Organization identity layer with knowsAbout, template schema architecture, entity homes, validation with before/after evidence. Three to four weeks.
Start with thisbilingual sprint · EN + ES
€4,400
Everything in the sprint, delivered across English and Spanish properties with one unified entity identity, correct inLanguage and hreflang coherence. About one week longer.
Start bilingualBoth include a thirty-day post-handoff window for any validation questions that arise. If an audit preceded the sprint, its findings scope the work and its prompt set becomes the before/after benchmark — the two are designed to chain. What the price does not hide: implementation is included, not quoted separately afterward, and there is no retainer attached. If ongoing monitoring or earned-media work makes sense after the sprint, that is a separate conversation with its own published pricing, never a surprise line item on this one.
Yes — arguably more than before, with one big caveat: the reason changed. Google retired FAQ rich results on May 7, 2026, and How-To rich results before that, so schema implemented purely to win SERP dropdowns no longer pays. But the same March 2026 update that narrowed display features saw sites with clean entity schema improve citation rates in Google’s AI answers, and Google and Microsoft have both confirmed structured data helps their AI systems understand content. Schema’s job moved from decoration to verification.
It is structured data that tells machines, unambiguously, who you are: your organization, what it does, the topics it genuinely covers, the services and products it offers, and how all of that connects. When an AI engine considers citing you, this is the machine-readable identity it checks against. Clean entity schema is the difference between a model naming you with confidence and hedging with a generic answer.
knowsAbout is an Organization property that declares the subject areas your company and authors genuinely have expertise in. Since Google’s March 2026 update it has emerged as one of the highest-impact entity signals, because AI systems use topical authority when selecting sources for a query category. Declaring the topics you actually cover, and only those, makes you more likely to be cited for queries in those domains.
Usually no. Both remain valid schema.org types: they cause no errors, carry no penalty, and Bing plus the AI crawlers still parse them. What we do recommend is an audit for markup that no longer matches visible content, schema added to non-primary content purely for display features, and anything Google’s March 2026 update now treats as noise. Keep what describes real content; remove what was a display trick.
Three to four weeks for a typical B2B SaaS site: one week of entity audit and architecture, one to two weeks of implementation across templates, and a validation week with before-and-after checks. Bilingual engagements run about a week longer because we unify the entity identity across both languages rather than duplicating it.
Yes, natively. We build and audit entity schema in English and Spanish, keeping one unified organization identity across both, with correct inLanguage values and hreflang coherence. For brands serving the US plus Latin America or Spain, a unified bilingual entity is one of the cheapest authority advantages available, because most competitors handle the second language as an afterthought.
The free AI visibility snapshot shows how five engines currently describe your brand — including the ambiguity and errors entity work fixes. It is the honest way to find out whether you need this sprint before paying for it. Forty-eight hours, async, no sales call: if your entity is already clean, the snapshot will say so and we will both have saved a month.
