In 2023, Rick Chorney started a cleaning company in British Columbia and did $242,000. Year two: just under $1 million. Year three: $1.3 million projected. He told Fortune he spent four hours setting up AI tools — an AI receptionist, automated intake forms, follow-up sequences — and went home at 8 PM instead of midnight.
Those tools worked because Chorney gave his AI the context it needed. On June 12, 2026, Google Cloud published the formal standard for structuring that kind of context: the Open Knowledge Format (OKF). This post explains what OKF is, where it came from, and what a contractor or service business owner actually does with it — no developer required.
What those AI tools did: they answered calls, booked jobs, and sent post-service follow-up. They did it correctly because someone had written down how the business worked — pricing, service area, booking rules, job types — in a format the AI could read and act on. The AI didn't guess. It knew.
Google's OKF is the standard that formalizes how to build that.
In short: Google's Open Knowledge Format (OKF) is a vendor-neutral standard for organizing your business knowledge — pricing, service area, job types, policies — into structured markdown files that AI agents can read and act on. For a plumber, roofer, or HVAC company, it's the difference between an AI that guesses and an AI that knows. No code required. A minimum viable bundle is five files and an afternoon.
What is Google's Open Knowledge Format (OKF)?
Google's Open Knowledge Format (OKF) is an open specification — not a product you pay for — that defines how to organize business knowledge into a folder of plain markdown files that AI agents can read and act on. Google Cloud announced OKF v0.1 on June 12, 2026, authored by Sam McVeety and Amir Hormati of Google's Data Cloud engineering team.
The official definition from the Google Cloud announcement: "OKF is a vendor-neutral, agent- and human-friendly standard for representing the metadata, context, and curated knowledge that modern AI systems need" (Sam McVeety and Amir Hormati, Google Cloud, June 12, 2026).
In practice, OKF v0.1 represents knowledge as a directory of markdown files with YAML frontmatter — short structured headers at the top of each file that classify what it is and what it covers. The content is whatever the AI needs to know about that topic, written as clearly as possible.
"Vendor-neutral" matters here. OKF isn't a Google Cloud feature you enable or a subscription you buy. It's a format convention — like PDF or HTML. You write OKF bundles in any text editor. They live in any git repo, any shared drive, any Notion workspace. Any AI tool that accepts context files can read them. The spec is public. You don't need Google to use it.
For a contractor, the simplest mental model: an OKF bundle is a folder of training documents for your AI tools. The AI reads the folder before it acts. What's in the folder determines how well it acts.
Where did OKF actually come from?
Google formalized a pattern that developers had already been discovering independently. In April 2026, Andrej Karpathy — co-founder of OpenAI, former Director of AI at Tesla — published a GitHub gist describing a system he called an "LLM wiki": a structured folder of markdown files that an AI agent reads, maintains, and references across all its work. The gist collected over 5,000 stars in its first week.
Karpathy's core observation was about where AI agents fail in practice. They can reason well. They can write, edit, and synthesize. What they can't do is remember, cross-reference, and keep information current across long time horizons — unless you give them a structured place to read from. His solution: build that place deliberately.
"LLMs don't get bored, don't forget to update a cross-reference, and can touch 15 files in one pass. The wiki stays maintained because the cost of maintenance is near zero" (Andrej Karpathy, GitHub, April 4, 2026).
Developers built variations immediately: AGENTS.md files for coding agents, CLAUDE.md files for Claude Code, Obsidian-to-agent pipelines for personal knowledge management. The pattern worked well enough that Google standardized it. OKF is Karpathy's pattern made portable, with agreed-upon field names so a bundle written for one AI tool can be read by another without translation.
The blog you're reading right now runs on this pattern. The CLAUDE.md file that governs this site's AI-assisted production workflows is a working OKF bundle: brand rules, design tokens, component conventions, blog doctrine, and current site status — all in one structured folder that Claude Code reads before it does anything on this codebase. It's not a demo or a test. It's the production system.
What problem does OKF solve for a service business?
The problem OKF solves is context fragmentation: your business's knowledge lives in your head, in old email threads, in an outdated pricing sheet, and in the unwritten rules only you know. An AI deployed without structured context guesses. An AI given an OKF bundle knows.
Consider what an AI receptionist needs to answer calls correctly:
- Is this caller inside your service area?
- What's the right price range for this type of job?
- Is this an emergency that warrants interrupting a job in progress?
- What does a new client need to provide to get scheduled?
Without a structured context file, the AI either hallucinates answers, asks the caller to repeat themselves, or routes every call back to you — which defeats the purpose. With an OKF bundle, it has the same information a well-trained new employee would have on day one.
The core failure mode of AI tools is not that they can't reason — it's that they reason from incomplete information and produce confident wrong answers. An AI that doesn't know your service area tells a caller you serve their zip code when you don't. An AI that doesn't know your emergency policy books a routine quote for tomorrow when the caller needed someone there today. An AI that doesn't know your minimum job size fills your calendar with jobs you'd turn down on the phone.
Each wrong answer costs more than a no answer. A caller who gets incorrect information from your AI doesn't just hang up — they leave with false expectations, call back frustrated, or leave a review about something your AI said. OKF exists specifically to close this gap: structured knowledge in, accurate behavior out.
What does an OKF bundle actually look like?
An OKF bundle is a folder of plain text files, each covering one topic your AI needs to know about. Files are written in markdown and start with a short YAML header classifying the content. The body is whatever that topic requires — written the way you'd explain it to a new employee. No special software, no database, no code.
A basic OKF bundle for a residential roofing company:
/roofing-knowledge/
index.md — what this bundle is; how to use it
service-area.md — counties and zip codes served; hard exclusions
job-types.md — storm damage, new construction, repairs, inspections
pricing.md — estimate ranges by job type; minimum job size
emergency-policy.md — what counts as emergency; after-hours protocol
booking-rules.md — what info is needed to schedule; availability windows
faq.md — the 10 questions you answer every week
Each file opens with a header:
---
type: policy
title: Emergency Call Policy
description: How to categorize emergency calls and what happens next
---
Then the actual content — written plainly:
"A caller describing active water intrusion through the roof during rain, ceiling bubbling, or visible daylight through the deck is an emergency. Call back within 30 minutes regardless of current job status. A caller saying the roof looks old or they want to replace it before winter is a standard quote request — schedule the next available inspection slot."
That level of specificity is what separates an AI that handles calls correctly from one that makes enough mistakes to erode customer trust. The AI acts on what the bundle says. Vague context produces wrong actions.
The same folder can serve multiple AI tools at once. An AI receptionist, a booking agent, an estimating assistant, and a follow-up automation can all draw from the same files. When your pricing changes, update pricing.md once — every tool reading it updates automatically. No re-training. No re-configuration. This is the "vendor-neutral" design working in practice.
How do AI agents use an OKF bundle to run your business?
AI agents operate by reading context and taking action. Without an OKF bundle, the context is generic — the agent knows how to have a conversation but not how to have your conversation. With a bundle, it reads your policies before it responds, routes calls against your actual criteria, and acts within your real rules.
The applications for a service business are direct:
AI receptionist. Before answering, reads service-area.md to verify the caller's zip code. Reads job-types.md to categorize the job correctly. Reads emergency-policy.md to decide whether this interrupts a current job or gets a next-available slot. Reads booking-rules.md to confirm the appointment. The call gets handled correctly from start to finish.
Lead qualification. Reads your minimum job size from pricing.md. If a caller needs a $300 repair and you only take jobs over $2,500, the AI flags it for manual review instead of booking it onto your calendar. You stop spending time on jobs you'd never take.
Estimate follow-up. Reads pricing.md to give realistic ranges when a prospect asks "how much will this cost?" — without committing to a specific number you haven't seen the job for. Reads faq.md to handle objections. Sends follow-ups on whatever timing you've written into the bundle.
Forrest Zeisler, co-founder and CTO of Jobber — one of the primary scheduling platforms for home service contractors in North America — was direct about the direction this is heading: "No one's going to benefit more than the small blue-collar businesses from AI. For them, time is literally money" (Fortune, March 2026).
The bundle doesn't create new AI capabilities. It makes existing capabilities relevant to your specific business. Tools that now answer calls generically start answering them specifically.
What happens to service businesses that don't structure their knowledge for AI agents?
Businesses that skip the knowledge layer deploy AI tools that perform unreliably — wrong service area answers, miscategorized jobs, escalations that shouldn't need your time. The AI gets blamed for the outcome when the real problem was that nobody gave it the context to perform correctly.
The structural risk is longer-term. AI agents are increasingly used to find, evaluate, and recommend service providers. When a homeowner asks an AI assistant to find a licensed roofer who handles storm damage in their area, the AI queries businesses whose information is structured and legible. A company with an OKF bundle — service area, credentials, job types, pricing ranges — is retrievable and actionable. A company whose information is scattered across an out-of-date Google Business profile and a 2021 Facebook page is not.
This is the same shift that happened when search engines replaced the Yellow Pages. Businesses with crawlable, well-structured websites got found. Businesses that existed only in print became invisible — not because their work was worse, but because the information wasn't organized for the discovery system that replaced it.
This dynamic is now central to how AI SEO works. We covered the mechanics in our breakdown of generative engine optimization — the short version is that structured, citable, AI-readable content gets surfaced in AI Overviews and agent queries, while unstructured content gets skipped entirely. An OKF bundle is the business-knowledge equivalent of that optimization.
Rick Chorney's cleaning company didn't have enterprise software or proprietary technology. It had structured context and AI tools that could use it. The jump from $242K in year one to just under $1M in year two didn't require hiring, moving markets, or building a custom platform. It required writing down how the business works in a form AI agents could read and act on.
That's what OKF formalizes. The window to build it before competitors do is shorter than most business owners think.
How do you build an OKF bundle if you're not a developer?
You write it the same way you'd write a training guide for a new employee — except in markdown files organized in one folder instead of a binder on the shelf. No coding required. The minimum viable OKF bundle for a service business is five files: service area, job types, pricing guidelines, booking rules, and a FAQ. Most contractors can build all five in an afternoon.
Here's the process:
Step 1: Create a folder. Name it your-business-knowledge or [company]-okf. It can live in Google Drive, Notion, a local folder, or a GitHub repository. The format is just files — where they live is up to you.
Step 2: Write service-area.md first. This is the highest-leverage file and the most common source of AI mistakes when it's missing. List every zip code or county you serve. List every area you don't. List exceptions — you'll cross a county line for jobs over a certain dollar amount, you serve a specific zip for one service type but not another. Be specific. The AI acts on what you write here.
Step 3: Write job-types.md. One paragraph per job category. Include what you do and what you explicitly don't take: "we don't do residential flat roofs," "emergency plumbing only for customers already on the maintenance plan." The exclusions are as important as the inclusions — they're what prevent the AI from booking jobs you'd turn down anyway.
Step 4: Write pricing.md. Ranges, not exact numbers. "Furnace replacement: $3,500–$8,000 depending on unit size and ductwork condition. Annual maintenance: $150–$200. Emergency dispatch: $350 flat fee plus parts." Ranges give the AI enough to qualify a caller and set expectations without committing to a number before you've seen the job.
Step 5: Write booking-rules.md and faq.md. Booking rules: what information you need from a caller to confirm an appointment. FAQ: the ten questions you answer every week, written out with your actual answers — not the polished version, the real one.
That's a working OKF bundle. Point any AI tool that accepts context to the folder, and it acts with the knowledge of a trained first-day employee.
If building and maintaining this is a distraction from running the business, ClickWerxs AI Services builds the context layer first, wires the agents to it, and keeps the bundle current as the business changes.
Frequently Asked Questions
Is OKF a Google product I pay for?
No. OKF is a format specification — an open standard, like HTML or PDF. Google published the spec on June 12, 2026, but you don't pay Google to use it and you don't need Google's tools to build one. The format is public: any AI tool that accepts context files can work with an OKF bundle. The full spec is at cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing.
Do I need a developer to build an OKF bundle?
No. OKF files are plain text written in markdown — asterisks for bold, pound signs for headings, dashes for list items. If you can write in Google Docs, you can write an OKF bundle. The knowledge-writing requires no technical skill. Connecting the bundle to specific AI tools may require help depending on the platform, but the core content — service area, job types, pricing, booking rules — comes entirely from you.
How is OKF different from a website FAQ page?
A website FAQ is written for human readers who are browsing. An OKF bundle is written for AI agents that need to take action. The difference shows up in specificity: a website FAQ says "we serve the greater Denver area." An OKF bundle says "we serve zip codes 80201–80299 and Jefferson County; we do not serve Castle Rock or any area south of US-285." A human reader can ask a follow-up. An AI agent cannot — it acts on what the bundle says, so vague context produces vague (and often wrong) actions.
Can I use the same OKF bundle with multiple AI tools?
Yes — vendor-neutrality is one of OKF's core design principles. The same bundle works with an AI receptionist, a booking agent, a follow-up automation, and an estimating assistant simultaneously. Update pricing.md once when your rates change, and every tool reading that file updates automatically. No re-training, no re-configuration per tool. Write once, deploy everywhere.
How often does an OKF bundle need to be updated?
When something changes: new service area, updated pricing, added or removed job types, changed booking policies. Most bundles need a meaningful update once or twice a quarter. The signal is usually a misrouted call or an AI giving a wrong answer — which tells you a specific file needs a more precise rule. The AI surfaces the gaps; you close them.
Kaleb Dickhaut — Founder, ClickWerxs. Kaleb built ClickWerxs from the ground up — from payment processing ISO to the Command Center platform to the AI SEO methodology the blog runs on. He works with contractors and service businesses on payment processing, CRM, and AI-enabled operations. linkedin.com/in/kaleb-dickhaut
Sources
- StatCounter Global Stats, search engine market share worldwide — Google 91.25% as of June 2026. gs.statcounter.com
- ClickWerxs blog, first-party AI citation data, May 2026 — one post generated 932+ Bing AI citations across query variants from a new domain; daily citations rose from 0 to a peak of 171 on 7 May 2026, settling at 56–111 per day, with Google organic clicks near zero over the same period. Reported as a past result, not a promise of performance. Operator data.
ClickWerxs sells SEO and AI visibility services and earns revenue from those engagements. First-party figures are past results for this blog and are not a promise of future performance. This is operator opinion, not professional advice.