Google's featured snippet, People Also Ask, and AI Overview all reward pages that answer the right questions in a structured way. The problem is figuring out which questions are the right ones. Reading ten competitor pages for a single query to compile that list by hand is tedious, and the list you come away with is never as comprehensive as it should be.
The FAQ Harvester does that reading for you. You paste a search term and the top ten URLs; it fetches each page, pulls every question it can find — from FAQPage schema, from <details>/<summary> accordions, from question-shaped H2 and H3 headings — deduplicates across sources, and hands you a consolidated list plus a ready-to-paste FAQPage JSON-LD block with the ten most common questions prefilled.
Why This Matters More Than It Used To
Ten years ago, FAQ sections were nice to have. Today, they are one of the few remaining levers a content writer has in a SERP increasingly dominated by Google's own features.
The AI Overview pulls directly from the questions and answers on ranking pages. The People Also Ask accordion is populated from the same source. Featured snippets lean heavily on Q/A structure. Schema-marked FAQPage content gets dedicated real estate on the SERP and dedicated inclusion in AI model training data. If the top ten pages for your query all answer the same eight questions and your page answers three of them, you've handed those five other questions to your competitors.
The Harvester finds the eight. Your job is to answer them better.
How It Extracts Questions
The tool runs three extractors against every page, in priority order.
FAQPage JSON-LD is the cleanest source. Every Question entity in the mainEntity array becomes an entry with the question text, the accepted answer text, the source URL, and a schema tag.
<details>/<summary> is the HTML5 disclosure element most accessible accordions use. The summary becomes the question, the details body becomes the answer. Extracted entries are tagged details.
Question-shaped headings is the fallback for pages that use accordions without schema. The tool scans H2/H3/H4/DT/strong/b elements, keeps ones that end in ? or start with how/what/why/can/is/are/do/does/should/when/where/which/who/will, then looks at the next 1–3 sibling paragraphs as the answer. Entries are tagged by the element type they came from.
The tagging matters because it tells you the source quality. Schema-tagged answers are verbatim from the competitor's structured data. Heading-based answers are my best guess at which paragraph answers the question — sometimes wrong.
Deduplication and Clustering
The same question gets phrased five different ways across ten pages. The Harvester normalizes each question (lowercase, strip punctuation), then pairs them with Jaccard similarity on the word set. Pairs above 75% similarity get merged into a single entry with a combined source list.
After dedup, the tool clusters the remaining questions by leading interrogative plus the first noun token. "How much does it cost" and "How much is" land in the same cluster; "What is" and "Why does" land in different ones. The Clusters tab surfaces these groupings so you can see which coverage themes matter most.
The JSON-LD Output
The JSON-LD tab builds a FAQPage block containing the top ten questions by source frequency — the ones asked by the most competitor pages. Each question includes a prefilled answer drawn from the extraction. You paste it into your page <head>, then replace the prefilled answers with your own authoritative, factual responses. The prefilled answers are a placeholder, not a finished product.
If you want a proper content refresh, the AI Enrichment Prompt tab takes the top ten questions plus their observed contexts and generates an LLM prompt asking for tight, authoritative 2–4 sentence answers written in your voice. Paste that into Claude or ChatGPT, edit the output, and you have a finished FAQ section.
When to Run It
When you're building a new page targeting a specific query. Use the Harvester before the first draft so the FAQ section exists from day one, not as a later patch.
When you have a page on page two of Google. Your existing page probably answers some of the expected questions. The Harvester tells you which ones you're missing.
When an AI Overview appears for a query you wanted to rank for. The AI pulls from Q/A content. Get your questions and answers into the same semantic space as the ranking pages.
When you audit a client site for the first time. Running the Harvester on their top 3–5 priority queries produces a content-gap deliverable in an afternoon.
Honest Limits
The Harvester only sees static HTML. A competitor who renders FAQs in JavaScript after first paint will look empty to the tool — which is still useful information, because those FAQs aren't indexable either.
The dedup threshold is fixed at 75% Jaccard similarity. Edge cases — two questions that share most words but mean different things — will occasionally merge. Review the deduped list if the output feels thin.
Answer extraction from non-schema sources is approximate. Pages that interleave marketing copy and answers will produce answer text that's half-promotional. Treat extracted answers as context, not as something to publish.
How This Fits the Methodology
The $20 Agency book has a full chapter on why the FAQ page is the most underused SEO real estate (Chapter 7). This tool is the automated version of the work it teaches: extract the right questions first, answer them second, mark them up with schema third. The same FAQPage JSON-LD block plugs straight into the Chapter 5 schema workflow. If you're scaling this across a network of sites, the $100 Network LLM-optimized content chapter (16) explains why consistent Q/A structure is the single strongest signal for AI Overview inclusion. For a brand-new site, $97 Launch Chapter 23 covers how to prioritize which FAQ questions matter most when you're starting from zero.