Cloudflare’s AI Search product moved out of beta and into general availability this week, and billing for it starts November 1, 2026. Cloudflare’s own announcement covers the technical upgrades in detail, multimodal image embeddings, OCR for scanned PDFs, larger file size limits. A small business owner evaluating this needs to know what this costs to run against a normal website, and whether it is worth it at that scale. The announcement has the pricing numbers, but nobody has run them against a realistic small-site workload yet.
Table of Contents
Breaking Down How Cloudflare AI Search Works
Strip away the product name and AI Search is a managed retrieval system. You point it at your content, text, PDFs, images, and it handles chunking that content, generating embeddings, indexing it, and serving search results, including letting external AI agents query your site’s content directly rather than just crawling it the way a traditional search engine bot does. The general availability release adds multimodal capability, meaning the system can now match an image search based on visual characteristics, texture, composition, and color, rather than only matching against a text caption generated from that image.
File size limits doubled from 4 MiB to 10 MiB, and OCR now automatically extracts text from scanned PDFs that previously would have been invisible to the system entirely.
The Pricing Structure
Three things get billed separately, and understanding each one matters because they scale very differently depending on your site. Content ingestion costs $0.75 per million tokens, with a 5 million token free monthly allotment included on all Workers plans. Stored data costs $2.00 per GB-month, with 10 GB included free. Semantic search queries cost $0.75 per 1,000 queries, with 1,000 free semantic queries included monthly. Ingestion is a one-time-ish cost per piece of content, assuming you are not re-indexing constantly. Storage is ongoing, billed every month regardless of traffic. Queries scale directly with how much search activity your integration generates.
Running the Numbers for a Small Business Site
Take a mid-sized small business site: a few hundred pages of content, call it 300 pages averaging 800 words each, which works out to roughly 240,000 words or somewhere around 320,000 tokens once you account for how tokenization typically runs a bit higher than a raw word count. That entire site’s initial ingestion falls comfortably inside the 5 million token free monthly allotment, meaning the one-time cost of indexing it in the first place is zero under the free tier, assuming you do not need to re-ingest the whole site repeatedly in a single month.
Storage is where a tiny site and a large one start to look different. Three hundred pages of text content is well under a gigabyte in raw size, meaning storage costs land inside the free 10 GB tier too, with room to spare even if you later add a meaningful library of product images or PDFs. A site built primarily around a large media library, lots of high-resolution images or long PDF documents, is the profile that will start accruing storage cost, since that is the dimension media-heavy content affects directly.
A product catalog storing a thousand high-resolution images at a few megabytes each could reach that 10 GB threshold on images alone, well before text content becomes a meaningful factor in the calculation.
Estimating Query Volume
Queries depend entirely on usage, which is the one variable in this calculation that is unpredictable in advance. A small business site that enables this mainly so AI agents and assistants can search its content more accurately might see a modest trickle of queries well within the 1,000 free monthly semantic query allotment. A business building an AI-powered search feature into its own site, something customers interact with directly and frequently, could exceed that free allotment within the first month. At that point the cost is still modest, $0.75 per additional thousand queries, but no longer zero.
Running a quick estimate before committing to a customer-facing feature, based on current site search volume if you already track it, is a reliable planning step rather than assuming the free tier will cover whatever usage pattern eventually emerges.
When Paid Tiers Make Sense
For most small business websites, the realistic read is that AI Search’s free tier comfortably covers the entire footprint of a normal content site: a few hundred pages, moderate query volume, and no unusual media-heavy content. The scenario where this becomes a line-item cost worth budgeting for is a site that either has a large, growing media library, such as an e-commerce catalog with thousands of product images and specification PDFs, or a business building an AI-assisted search experience directly into its site for customers, generating sustained query volume past the free allotment every month.
That second scenario is also where the value proposition is strongest. A business paying money for AI Search queries is doing so because it has built a useful AI-powered search or support experience that customers are actively using, which is a different cost-benefit calculation than a brochure site turning the feature on just because it exists.
Implications for Non-Builders
If your interest in this is about AI agent discoverability, the idea that tools like ChatGPT or Perplexity might index and cite your content more accurately, rather than about building a customer-facing search box, the free tier is likely all you will ever need. Enabling AI Search primarily to make your existing content more legible to external AI agents is a low-cost, low-risk move for almost any small site, since the ingestion and storage costs that matter at scale do not show up until you are past what a typical content site generates.
The Bottom Line
Cloudflare’s own announcement correctly emphasizes that the managed pricing model removes upfront capacity planning, and running the numbers against a realistic small-site workload backs that up: a typical few-hundred-page business site comfortably fits inside the free allotments across all three billing dimensions. The cost conversation only becomes real once you are either running a large media library or deliberately building a high-traffic, customer-facing AI search feature. In either of those cases, the per-unit pricing is modest enough that the decision is about whether the feature is worth building at all, not whether this specific pricing model makes it prohibitively expensive.
Related reading: How Much Bandwidth Does Your Website Actually Need and Google Search Console Can Now Show You Image-Based Search Traffic.