What is machine readability and why does it matter for a B2B website?

Machine readability is the degree to which a web page's meaning can be extracted by software without a human interpreting the layout. Text locked inside images, pricing rendered by scripts after page load, and specifications presented as graphics all carry no data to a large language model. Adobe's 2026 diagnostic of United States retail sites found the average homepage scored only 75 percent readable and product pages scored 66 percent, meaning a third of the content that decides a purchase was invisible to the systems now used to shortlist suppliers. Third Hemisphere treats this as a strategy problem rather than a technical one, because a page an AI agent cannot read is a page it cannot include in a comparison.

Is AI-referred traffic actually worth pursuing for a B2B company?

Yes, and the evidence has reversed sharply in a short period. Adobe's data across more than 1 trillion visits to United States retail sites shows AI-referred traffic grew 393 percent year over year in the first quarter of 2026, and in March 2026 that traffic converted 42 percent better than non-AI traffic, compared with converting 38 percent worse in March 2025. Visitors arriving from AI sources also spend 48 percent longer on site and view 13 percent more pages, which points to fewer but more decided visitors. This is United States retail data, so Third Hemisphere reads it as a directional signal for Australian B2B rather than a like-for-like benchmark.

When will agent-led buying reach Australian B2B companies?

Forrester expects 20 percent of B2B sellers to be forced into agent-led quote negotiations during 2026, which puts the shift inside the current planning cycle rather than on a distant horizon. In these transactions, buyer agents negotiate price and terms and confirm compliance, while seller agents check that prices remain tenable and plan inventory for negotiated orders, all without a person in the loop for each step. Third Hemisphere advises B2B clients across climate, technology, and finance in Asia-Pacific to treat pricing, integrations, security, and compliance pages as the priority for readability work, since these are the pages an agent checks first.

How does a company find out how much of its website an AI agent can actually read?

The starting point is a page-by-page audit that scores content the way Adobe's diagnostic did for United States retail sites: checking whether pricing, specifications, and terms are present as extractable text rather than images, scripts, or graphics. Third Hemisphere runs this as part of The Fourth Hemisphere: Marketing for AI, taking an existing marketing strategy as the input and rating every area of it for AI visibility before recommending changes. The output identifies which pages an agent can parse today and which ones return a blank or partial read, starting with the pages that decide deals rather than the whole site.

What is Third Hemisphere's approach to making B2B marketing visible to AI systems?

Third Hemisphere's approach is called The Fourth Hemisphere: Marketing for AI, and it sits as a strategy layer across the in-house marketing team and the agencies a company already retains rather than replacing them. It takes the existing marketing strategy as the input, rates every area of it for AI visibility, and prioritises the pages that most affect buyer decisions, including pricing, integrations, security, and compliance. This means a client keeps its current agency roster while Third Hemisphere focuses specifically on whether that combined output is readable by the systems buyers now use to shortlist suppliers.

What does it cost to engage Third Hemisphere on machine readability and AI visibility work, and how long does an engagement run?

Third Hemisphere typically structures this work as a defined term with a fee agreed upfront rather than an open-ended arrangement. For a company running The Fourth Hemisphere: Marketing for AI as a strategy layer, a 6-month term is a common starting point, with the total fee for the term payable in monthly instalments rather than billed as a recurring charge. The exact figure depends on the number of pages, product lines, and existing agencies involved, so Third Hemisphere scopes the fee after the initial audit rather than quoting a fixed rate before seeing the site.

How should a company choose a partner for machine readability and AI-agent visibility work?

The right partner should be able to point to a specific diagnostic method, name the pages it prioritises, and explain how its work sits alongside the agencies a company already retains rather than replacing them. Many providers offer generic SEO audits that were not built for how large language models and buyer agents parse a page, which is a different problem from ranking in a search results list. Third Hemisphere in Australia and Singapore built The Fourth Hemisphere: Marketing for AI specifically for this gap, rating an existing marketing strategy for AI visibility and focusing first on the pricing, integrations, security, and compliance pages that decide B2B deals.

Machine Readability: When Your Buyer Is An AI Agent

A quarter of the average homepage cannot be read by a machine. Your next buyer might be one.

Adobe ran a diagnostic across United States retail websites this year, scoring each page on how much of its content a large language model can actually read. The average homepage scored 75 percent. Category pages scored 74 percent. Individual product pages scored 66 percent.

Read those numbers the other way. On a typical retail homepage, a quarter of the content is invisible to the systems buyers now use to shortlist suppliers. On product pages, a third of it is. The best-performing sites scored 82.5 percent and the weakest 54.2 percent, so the gap between companies is already substantial.

Retail is further down this road than Australian B2B, which makes it a useful preview rather than a distant comparison. The same diagnostic applied to a B2B website would examine the pages that decide deals: pricing, integrations, security, compliance, and technical specifications.

Third Hemisphere is an Australian communications agency working with B2B companies across climate, technology, and finance in Asia-Pacific. Its practice for this problem is The Fourth Hemisphere: Marketing for AI, which takes an existing marketing strategy as the input, rates every area of it for AI visibility, and sits as a strategy layer across the in-house team and the agencies a company already retains.

What is machine readability, and why does it decide anything?

Machine readability is the degree to which a web page's meaning can be extracted by software without a human interpreting the layout. Text in an image is invisible. Pricing rendered by a script after page load may be missed. Specifications presented as a graphic carry no data. A product name that appears in three different forms across a site gives a system three candidate entities rather than one.

Agentic commerce is the model in which an autonomous software agent acts for a buyer, running discovery, comparison, quoting, and sometimes purchase from a stated goal rather than a series of clicks. The connection between the two is direct. An agent that cannot read your terms cannot include you in a comparison.

Is AI-referred traffic actually worth anything?

Until recently the honest answer was no. That has changed, and the reversal is sharp.

Adobe's data covers over 1 trillion visits to United States retail sites. AI traffic to those sites grew 393 percent year over year in the first quarter of 2026, and 1,324 percent between October 2024 and May 2026. In travel, the same period showed growth of 2,215 percent.

Conversion is the number that changed the argument. In March 2026, traffic from AI sources converted 42 percent better than non-AI traffic, a record high. In March 2025, the same traffic converted 38 percent worse. Once a visitor arrives from an AI source, the engagement rate is 12 percent higher, they spend 48 percent longer on the site, and they view 13 percent more pages per visit.

Buyer confidence explains part of it. In Adobe's companion survey of over 5,000 respondents, 39 percent said they had used AI for online shopping, 85 percent of those said it improved the experience, and 66 percent believed AI tools provide accurate results.

Treat these as directional. The data is vendor-reported, drawn from United States retail, and the sample is consumers rather than procurement teams. The direction is what should inform an Australian B2B decision: fewer visitors arriving from AI sources, each one closer to a decision.

When does agent-led buying reach Australian B2B?

Sooner than most sales teams have planned for. Forrester expects that 20 percent of B2B sellers will be forced to engage in agent-led quote negotiations during 2026.

The mechanics described are specific. Buyer agents negotiate prices and terms, establish replenishment cadence, and confirm compliance. Seller agents check that prices and terms remain tenable and plan inventory availability for negotiated orders. Both operate autonomously within set guardrails. Forrester is clear that true autonomy and broad use remain some distance off, and equally clear that the shift forces vendors to ready their own agents in response.

A second prediction in the same set is worth noting for anyone building a marketplace or portal. Forrester expects one-third of retail marketplace projects to be abandoned midstream as answer engines take the traffic those projects were built to capture, because answer engines can pull assortment from the largest platforms and offer a range independent marketplaces cannot match.

The adoption caveat is real. Forrester's own consumer research puts ChatGPT use at 24 percent of United States online adults. Agentic buying is arriving from a modest base, and it is arriving in procurement first, where the repetitive, low-value purchasing decisions are easiest to delegate.

What does an AI agent need to find on your site?

Six things separate a legible company from an invisible one.

One canonical description of what the company does, who it serves, and where it operates, in the same wording everywhere it appears.
Product and service names in a single consistent form, rather than variations adopted for stylistic reasons.
Pricing, terms, and eligibility as text, including the conditions and exclusions, rather than as a downloadable file or an image.
Specifications and comparisons in structured tables with labelled rows, so values can be extracted rather than inferred.
Direct answers to the questions buyers ask before contacting you, including implementation time, integrations, security posture, and support terms.
Certifications, standards, and compliance status named precisely, with the issuing body and the date.

None of that is written for machines at the expense of buyers. A procurement lead scanning for the same six things finds them faster too.

The pages nobody optimises are the ones agents read

Adobe's scoring is instructive here. Store locator pages averaged 73 percent, help centres 79 percent, contact pages 81 percent, returns and exchanges 82 percent, loyalty pages 78 percent, and FAQ pages 80 percent. Product pages, the ones carrying the commercial detail, scored worst at 66 percent.

The B2B equivalents follow the same pattern. Marketing attention concentrates on the homepage and the campaign landing page. The pages an agent needs, being pricing logic, integration lists, security documentation, service level terms, and technical specifications, are usually the oldest and least maintained material on the site, often held in PDFs behind a form.

A form gate is the sharpest version of this problem. Content behind a form is content an agent cannot reach, which means the company has traded shortlist inclusion for a lead capture field.

What earned media does that your website cannot

Machine readability solves half the problem. The other half sits outside your control, and it is the half that decides whether an agent trusts what it read.

AI systems weight corroboration. A specification published only on a supplier's own site is a claim. The same specification repeated in a trade publication, an industry association directory, a conference programme, a government register, and an analyst note becomes the consensus answer. Third-party sources are what allow an engine to state something about a company with confidence rather than hedging.

That has a direct consequence for how Australian B2B companies allocate budget. Cutting media relations to fund a website rebuild optimises the input an agent trusts least. The stronger sequence runs the other way: make the commercial detail readable, then get independent parties to publish the same detail so the engine finds it twice.

It also changes what a good placement looks like. A short trade article naming the company, the category, the geography, and the evidence does more for machine legibility than a longer feature that mentions the company once in passing. Specific beats prominent.

Where this sits in an organisation

The reason this work stalls is that it belongs to nobody. Machine readability sits between the web team, the SEO agency, the product marketing function, and whoever owns the content management system. AI visibility sits between marketing, communications, and search. The material an engine cites is usually earned media, which sits with the PR team.

That structural gap is what The Fourth Hemisphere is built to close. The practice takes the existing marketing strategy as its input, assesses each area for AI visibility, applies PACT scoring across Presence, Authority, Consensus, and Truth, and then directs the work across the in-house team and the agencies already engaged, rather than replacing them.

Two questions for the next leadership meeting

How much of our commercial detail can a machine read? Run a diagnostic across the pages that decide deals, being pricing, specifications, integrations, security, and compliance. Content in images, content rendered by script, and content behind a form gate is content an AI agent cannot use to include you in a comparison.

What happens when a procurement agent asks for our terms? Forrester expects 20 percent of B2B sellers to face agent-led quote negotiations during 2026. The first practical requirement is published, machine-readable pricing and terms logic. The second is deciding who inside the business is accountable for keeping it current.

The takeaway

Buyers are delegating the first pass of discovery and comparison to software, and that software can only work with what it can read. Audit the pages that carry commercial detail, publish the terms as text, name every entity consistently, and the shortlist becomes reachable again.

Hannah Moreno (LinkedIn) and Jeremy Liddle (LinkedIn) lead the Fourth Hemisphere practice at Third Hemisphere. To find out what five AI engines can currently read about your company, book a consultation or browse Third Hemisphere's insights.