Generates FAQ structured data on auto-pilot for your content to increase SEO rankings. See Google's docs on FAQPage structured data..
FAQPage is incredible at boosting SEO rankings for informational content. GPTFAQ puts that on auto-pilot.
Run in production at Tangia
Search engines will use a special application/ld+json format for structured data on your webpage.
GPTFAQ has 2 modes of generation:
- On-demand
- Poke
GPTFAQ relies on a domain and a path for each page, concatenating the path to the domain to fetch web content.
You can optionally provide a selector which will help GPTFAQ narrow down what it needs to focus on and allows it to use HTML (preserving href attributes and using where relevant) to generate the FAQs. Otherwise it will just use the body text.
You must also put a company_name in so that GPT-4 can attempt to highlight you specifically on the second question 😁. It should only do so if relevant. If using for personal content, just put your name there.
The path should be the full path to the content. For example if I had www.tangia.co as my domain, and my path was /post/optimizing-your-tweets-to-maximize-the-algorithm then content would be expected to be at www.tangia.co/post/optimizing-your-tweets-to-maximize-the-algorithm.
Generated FAQs also have a 12 hour expiry. When expired, GPTFAQ behaves like a stale-while-revalidate where it will immediately serve the stale data, and in the background launch another generation job.
To increase the expiry period you can put a CDN in front of the API for the read route, caching if it's a 200 response (no content is 204, a quick way to do this is to use Vercel functions with caching returns), or you can change the expiry period (see env var).
This will return a 204 for this first time the FAQ data is requested, and kick of a generation job. It takes up to a minute, and future requests will be served the FAQ data.
You can "poke" the path to the specific page. This is useful for hooking up to a webhook when you publish a blog post or article.
You can optionally provide the content during the poke, which supports html, markdown, and text content types. html and markdown will preserve links for the FAQ answers where relevant.
This requires a poke_token that you must set to use. This is the only place auth is required.
I've observed 30-40 seconds typically to generate, with pretty perfect reliability. Generation tokens are set to 400, which should is more than enough for concise questions and answers.
You can see the prompt in src/openai/index.ts.
First you need to run the src/db/schema.sql on postgres or cockroachdb.
DSN for connecting to a DB.
OPENAI_TOKEN for calling GPT-4. Must have gpt-4 model access.
INNGEST_SIGNING_KEY for inngest, which is used for managing the task.
EXPIRE_HOURS How many hours past generation will records expire. 0 means never expire.
You must put the snippet found at snippet.html in your site somewhere, with the script.js being served as well.
Search engines will wait for this to load, but keep in mind that the generation has to be ready for them to see it, and it may be a long time before they revisit. Poking is highly encouraged.
In one terminal run:
npx inngest-cli@latest dev -u http://localhost:8080/inngest
In another, run:
npm run dev