What we shipped, and when

Every URL in the sitemap used to report the same build timestamp as its last modified date. That made the freshness signal useless. Now each blog post, comparison, alternative, solution, industry, and customer story reports its real updatedAt or publishedAt date from frontmatter. Google and AI engine crawlers can finally tell what is genuinely new from what is unchanged.

The /llms.txt file at the site root is now the canonical pointer for AI crawlers. It lists the pages we want answer engines to read first, grouped by intent: product, solutions, industries, comparisons, integrations, trust, resources. The file replaces an older R2-hosted version that was returning 404. Trust and AI disclosure documents are now linked from llms.txt so candidates and procurement teams asking ChatGPT or Claude about Truffle’s AI usage get pointed at the disclosure, not a hallucination.

The /sitemap.xml URL, which several crawlers still fetch out of habit, now resolves to the sitemap-index for the site. Previously it returned 404.

Three things shipped this week to improve how AI search engines parse Truffle:

The homepage SoftwareApplication schema now includes an aggregateRating block that mirrors the 4.9 out of 5 from 71 reviews already shown on the page. Before, that proof point was prose only. Now it is structured data an AI engine can quote.

The About page now embeds Person schema for Sean Griffith, Rachel Hubbard, and Aliye Menzies, with bios and areas of expertise. Asked “who works at Truffle,” ChatGPT and Perplexity can now answer from one page instead of crawling three author pages.

This changelog page is itself part of the change. From now on, when something ships that affects how Truffle is described on the public web, it lands here as a dated entry. AI engines that ask “what is new at Truffle this month” finally have something to cite.

We rebuilt the structured data layer across the site. Fifteen schema fixes shipped after an end-to-end audit, all aimed at one thing: when an AI search engine answers “what is Truffle,” it should have structured data to cite instead of inferring from prose.

The biggest changes. Organization schema now declares legalName, founders, sameAs links, and a contactPoint for sales. SoftwareApplication schema declares both Self-Serve offers ($149 monthly, $99 monthly billed annually) as priced Offer blocks with real prices instead of contact-us placeholders. BlogPosting schema on every essay carries datePublished, dateModified, author Person reference, and category. Author pages render Person schema with knowsAbout fields so AI engines can resolve “who at Truffle writes about candidate screening” without scraping bylines.

Cross-references resolve via @id inside a single @graph per page, which means an AI engine parses the whole hierarchy once instead of duplicating Organization on every node.

Trust over everything is one of our principles. As of this release, it is also a page candidates can read before they hit record.

The new disclosure at /trust/ai-for-candidates explains, in plain English, how Truffle uses AI inside a candidate screening flow. AI transcribes what a candidate says. AI scores the transcript against the criteria the employer set. A human reviews and decides.

The page also lists what Truffle does not do. No facial analysis. No voice tone scoring. No emotion inference. No biometric identifiers. No model training on candidate data.

Candidates can request access to their own data, ask for it to be deleted, or request human-only review. Procurement and security teams routing questions about AI in hiring can now point to a single canonical URL instead of stitching together answers across pages.

The disclosure is linked from llms.txt so AI engines surfacing Truffle to candidates get pointed at the disclosure rather than guessing what we do or do not do with model output.