FAQPage and HowTo schema.
FAQPage translates a page’s question-and-answer pairs into machine language; HowTo does the same for the step-by-step order of a task.
These two types hold a special place in answer engine optimization because both work directly with the unit of the “question.” When an answer engine composes a reply, this is exactly what it looks for: a short, complete answer matching the question. Marking up the question-answer pairs already on the page spares the model from inferring the match from the text — it reads it explicitly.
The difference between the two lies in the shape of the answer. The FAQ schema is for independent questions; each answer is complete in itself and the order does not matter. The how-to schema is for a sequential procedure: the steps follow one another, and the order is part of the information. Choosing the wrong type means describing the structure wrongly.
How to use the FAQ schema
The usage rule is simple but frequently violated: every question and answer in the schema must also be VISIBLE on the page. A question-answer block placed only in the schema, with no counterpart on screen, is both an explicit violation of search engine rules and a half-measure for the user. The model quotes the answer; the visitor arrives at the page and cannot find it.
The second rule is that the questions must be questions people actually ask. Questions invented to place keywords — like “what is the best X service” — help no user and read as artificial. A good question list comes from real customer correspondence: what do people ask when requesting a quote, what worry do they voice, what do they wrongly assume.
The third rule is the form of the answer. The first sentence should answer the question alone and without context; detail comes after. This is the basic rule of citability, and it is especially decisive in FAQ blocks because the model already treats the block as a unit.
- Let the question be one actually asked — mine it from customer correspondence.
- Let the answer’s first sentence stand correct on its own — no back-references.
- Let every pair in the schema also appear on screen — no exceptions.
- Keep the answer short — two or three sentences; a long answer does not get quoted.
- A reasonable number of questions per page — a thirty-question list can give depth to none of them.
The how-to schema
This type describes a sequential procedure: which steps, in which order, with which tools, and how long it takes. Answer engines can use this structure directly when answering “how to” questions, because the steps are already separated into units.
The most important design decision is whether the steps are genuinely sequential. A list whose order does not matter — say, the items in a service’s scope — is not a how-to and must not be marked as one. The wrong type choice leads the model to misread the structure, asserting an order that does not exist.
On this site the how-to schema is used on the service pages, and the reasoning is measured: the service process really is sequential — discovery, art direction, build, measurement, and handover. The steps stand on screen in the same order under the same names; the schema translates them into machine language and invents nothing.
A list whose order does not matter is not a how-to; marking it as one misdescribes the structure.
The rich result expectation
Whether these schemas produce a visual rich result in search has changed over time, and Google has moved to narrow those displays. Adding the schema just to see an expanded box in the results is therefore now a weak justification.
The strong justification is different: the schema tells machines explicitly which questions the page answers, and that eases the matching work of answer engines. Whether or not a rich result shows, this benefit remains. It is more accurate to treat the schema as a readability layer than as a visual reward.
The practical consequence: put the question-answer blocks on the page for the user first, then mark them up. When the order is reversed — schema first, then something bolted onto the screen — the result is both an artificial page and a rules-violation risk.
Where to find the questions
The quality of the question list matters far more than the quality of the schema, and a good list is collected, not invented. The richest source is existing customer correspondence: what does someone requesting a quote ask, which worry do they voice, what do they wrongly assume. These questions are written in real language, and they are searched in that language too.
The second source is the query report in Search Console: which question-shaped queries bring people to your site? The third is the “people also ask” boxes in search results — they show directly what people wonder about the same topic. The fourth is the questions asked on your field’s community platforms; the language there is especially valuable because it has not been scrubbed into marketing speak.
When choosing among the collected questions, the criterion is: can I genuinely answer this, and can the answer be kept short? A question whose answer runs a full page belongs on its own page, not in an FAQ block. The block is for short-answer questions.
Maintenance: questions go stale
FAQ blocks written once and forgotten go stale faster than other content, because they are tied directly to current facts. Answers like “how long does it take,” “which technology do you build with,” “which languages do you support” turn false as the business changes — and a wrong FAQ is worse than none, because it misleads the user and, through the schema, announces the error to machines.
The practical solution is keeping the blocks in a source separate from the templates and reviewing them regularly. On this site the questions live in their own strings, independent of the page template; updating an answer means changing a single line. A question list embedded in a template is both hard to update and a doubled burden when translation comes.
One last measurement note: the most direct way to see whether FAQ blocks are working is to watch whether those questions themselves start appearing in Search Console’s query report. If, after adding a block, you receive impressions on queries shaped like the question, the match is forming; if nothing shows, either the question is not really asked or the answer is not direct enough. Both are fixable — but only after being measured.