FAR.AI: a research site rebuilt around the people who use it

Overview

• Client: FAR.AI, an AI security research and adoption non-profit
• Work: a Blueprint Audit, then a full migration and rebuild of far.ai from a Netlify deployment pipeline to native Webflow hosting, with an ongoing Monthly Partnership
• Outcome: immediate publishing, a publications library that finds things, structured data that validates clean, and search click-through rate up from 1.1% to 2.4%

Who the site is for

FAR.AI's site serves three audiences: people seeking impactful work in AI, who need to understand what a role here involves; the field building community, who need to know what an event covered or will cover; and readers of the research, technical and not, who need the blog posts alongside the papers to make the work legible.

The traffic bears that out. Careers is the busiest page on the site, team and individual researcher profiles rank next, and most non-brand search traffic is people looking up FAR.AI researchers by name.

The problem, in FAR.AI's words

At intake, the team described a site with content going out constantly and a workflow adding friction at every step.

The site ran on Webflow, but behind a pipeline that scraped the published output and redeployed it through Netlify. Every CMS change needed a separate deploy before it went live, crawlers were served through a pre-rendering layer, and the export had broken the 404 page. Events pages, edited several times a week, carried the heaviest load on that workflow.

The research section had its own problems. Searching the publications library for an author's name returned nothing. The Authors filter was empty. Keyword search only matched titles, so a paper about fine-tuning, for example, couldn't be found by searching for it. Research agendas hadn't kept pace with new publications and were difficult to navigate to. Links shared on Slack or LinkedIn came through without a proper preview. And underneath it all sat a long tail of one-off bugs.

Asked what success would look like six months on, the team named four things: people understand what FAR.AI does, the team has statistics it can rely on, there's a clear process for putting out content, and everything works when they link to it.

What Socialectric did

The Blueprint Audit assessed performance, accessibility, search visibility, readability to AI tools, and the content architecture, then set a phased plan. Four fixes shipped in the first month, along with consolidating a duplicated analytics setup into a single reliable property.

The rebuild that followed moved the entire site off the deployment pipeline and onto native Webflow hosting:

• A CMS built around how the team actually works. Speakers, paper authors and team members used to live in three separate collections, so the same person could exist three times over, retyped by hand each time. Now there is one collection, called Authors. Events reference their speakers from it, publications reference their authors from it, and adding either is a lookup rather than a new biography. Five hundred records were merged into it during migration.
• A publications library that finds things. Search now covers titles, abstracts and body text. The author filter is populated from the Team Members collection. Topics are multi-select tags, so a paper can sit under everything it's relevant to.
• One component framework across every page. Headings, cards, event blocks and profile layouts were rebuilt as a single shared component set, with fluid typography so type scales with the viewport rather than jumping between breakpoints. A change made once is pushed everywhere it is used, and new pages get built from parts that already exist. The site holds its shape as the team adds to it.
• A redirect map built from analytics and crawl logs, so existing links to research, press coverage and job postings kept resolving through the URL changes.
• Accessibility fixes across navigation, links and interactive controls, resolving the failures the audit surfaced.
• Metadata and structured data across the site. Page titles, descriptions and share previews for every page and template, JSON-LD schema aligned to the CMS fields in use, Wikidata entities for the organisation and its key researchers, and an llms.txt file describing the site to AI tools.

Results

Publishing is immediate. Events, news, research and profiles go live from the CMS the moment they're published. No deploy step, no pre-rendering layer, and a 404 page that works.

The site is now machine-readable for what it's for. Structured data validates clean across every type Google reports on for the site: events, profile pages, video and breadcrumbs. Events carry their dates, locations and speakers into search rather than sitting as plain pages. Event recordings are marked up as video. Researcher profiles are structured as people, which is what most non-brand search for FAR.AI actually is: someone looking up a researcher by name.

Search click-through rate more than doubled, from 1.1% to 2.4%. Daily clicks held steady on roughly half the impressions: a smaller, higher-intent search footprint doing the same work.

The publications library answers the questions researchers ask it. Author names return results. Keyword search reaches the abstract and body. Topic filtering works.

Links work when shared. Every page and template carries proper title, description and preview metadata, one of the four success criteria the team set at intake.

AI assistants are now a measured acquisition channel. Referrals from tools like ChatGPT and Claude show up as tracked traffic, and the site gives those tools an accurate account of FAR.AI rather than leaving them to infer it.

Accessibility failures on the pages that matter most are gone. The audit found 15 serious issues on the homepage alone, mostly links with no accessible name and nested interactive controls.

The team has statistics it can trust. Analytics consolidated to a single property with consent handling in place, another of the team's own success criteria.

Since then

The engagement continues past launch. The Monthly Partnership covers ongoing monitoring of performance, search visibility, accessibility and broken links, with issues fixed as they surface. Analytics get reviewed against what each part of the site is there to do.

An example from the first weeks after launch: FAR.AI noticed some images loading larger than their display size. The root cause was a Webflow limitation, where images brought in through CSV import don't get responsive variants generated. I applied the fix across every CMS image on the site, not just the handful that had been spotted.

The day-to-day work runs alongside FAR.AI's comms team. They bring what the organisation needs next, whether it's a new programme page, an event format, or a section that doesn't exist yet, and I fit it into the existing component system. The site expands without the team losing the ability to run it themselves.

A newsroom section is currently in design, and more projects are to come.