A B2B site that is merely published does not sell

Netsujo Inc. · Published August 3, 2026 · Updated August 3, 2026

We audited the public websites of 21 Japanese B2B technology and professional-services companies against 75 checks covering everything from whether a page exists to whether the site can be improved. On HTTP reachability, 0 of 21 companies were not met — the sites open. On heading structure 21 of 21 companies, on FAQ coverage of purchase blockers 17 of 21, and on pricing, duration and conditions 12 of 21 fell below the full standard. Being published and being retrieved, compared and contacted are different problems.

Scope of this report: we did not estimate lost enquiries or lost revenue. What we audited is the breakage, observable from public information, where an opportunity can be lost.

"Below the standard" means partially met plus not met. Where we report the stricter figure — companies that were clearly not met — we say so explicitly. Companies we could not observe are removed from the denominator and never converted into failures.

01 / Four kinds of breakage, repeated across the sample

02 / The problems were chained, not isolated

We produced one opportunity trace per company (21 in total) and selected 5 for publication: each covers breakage in two or more of the eight stages, and no two come from the same sector. The table below shows how wide each chain was. The observed statements themselves are recorded in Japanese by the audit, so the full chains are published on the Japanese page.

The remaining traces are not published because another company in the same sector is already shown. We state this so that a shorter list is not read as a smaller problem. What we can say from public information is where the chain breaks; actual drop-off, lost enquiries and revenue effects are not observable from the outside.

Full opportunity traces, step by step（日本語）The Japanese page lists every observed statement in each chain, with its source finding ID.

03 / Six checks in detail

12 claims were approved for citation. Listing all of them at the same length would be unreadable, so these six — chosen to cross every stage of the sales path — are set out with the exact pass, partial and fail conditions, what we observed, and how to check the same thing on your own site.

Heading structure

Below the standard: 21 of 21 companies. "Below the standard" is the sum of partially met and not met. Counting only the companies that were clearly not met gives 1 of 21.

What we checked. Not font sizes or visual styling, but the HTML heading tags. For every captured page we counted whether there was exactly one H1, and whether any heading level was skipped (for example H2 followed directly by H4).

Fully met

Every captured page has exactly one H1 and no skipped heading levels.

Partially met

Some pages have zero or multiple H1 elements, or a skipped heading level.

Not met

Every captured page has a problem with the H1 count or the heading levels.

Removed from the denominator

Companies where none of the confirmed pages could be captured are not assessed and do not appear in the totals.

What we observed.

• Fully met — none of the 21 companies reached the full standard for this check.
• Partially met — 20 companies: some — but not all — captured pages had no single H1, or skipped a heading level.
• Not met — 1 company: every captured page had no single H1, or skipped a heading level.

With this in place, the following become harder for a buyer to establish.

• whether the page has a single subject
• whether each section sits under that subject or is a separate topic
• where one explanation starts and ends, for anything reading the page mechanically

How to check your own site.

• Open your main service page, right-click and choose "View page source".
• Search for "<h1" and count the ones that contain text. An H1 holding only a logo image counts as zero here.
• Read the headings in document order and check that none is two or more levels deeper than the one before it (H2 followed by H4, for example).
• Repeat on your top page, service page, pricing page, case study and contact page.

If you leave it: You are left unable to confirm whether the relationship between the page subject and its sections is readable by a machine. What this study cannot conclude: Any causal relationship between heading structure and ranking.

Displayed publication date

Below the standard: 20 of 21 companies. "Below the standard" is the sum of partially met and not met. Counting only the companies that were clearly not met gives 4 of 21.

What we checked. Only article and case-study pages were in scope, and only whether a Western-calendar year-month or year-month-day appears anywhere in the body text — not whether it is labelled as a publication or update date. Japanese era dates were not detected. A date present is not treated as evidence that the content is current.

Fully met

Every article and case-study page in scope carries a Western-calendar date anywhere in the body.

Partially met

Only some of the pages in scope carry a Western-calendar date.

Not met

None of the pages in scope carry a Western-calendar date.

Removed from the denominator

Companies with no article or case-study page are out of scope, and companies whose body text could not be captured are recorded as not observable. Both are removed from the denominator.

What we observed.

• Fully met — 1 of 21 companies: every article and case-study page carried a date.
• Partially met — 16 companies: only some article and case-study pages carried a date.
• Not met — 4 companies: no article or case-study page carried a date.

With this in place, the following become harder for a buyer to establish.

• when the published results and case studies date from
• whether the stated conditions still apply
• which of two pages is the newer one when they disagree

How to check your own site.

• Open three of your case-study pages and check whether a publication or update date is shown in the body.
• Where a date is shown, confirm with the owner that it matches when the content was last revised (needs an internal check).
• Check whether pages stating pricing, results or scope carry a date at all.
• Check whether any page shows its date only in the index listing and not on the page itself.

If you leave it: Readers are left unable to judge what point in time the published information refers to. This study does not assess whether the content is out of date. What this study cannot conclude: How current the content is. A date alone does not guarantee freshness.

FAQ coverage of purchase blockers

Below the standard: 17 of 21 companies. "Below the standard" is the sum of partially met and not met. Counting only the companies that were clearly not met gives 14 of 21.

What we checked. Whether words matching four topics — cost, duration, the adoption process and support — appear in the FAQ page text. This is a search across the whole page text including navigation and footer, so it does not establish that the answers address those topics, and the quality of the answers was not assessed.

Fully met

An FAQ page exists and covers all four topics.

Partially met

An FAQ page exists but covers only one to three of the topics.

Not met

No FAQ was found among the captured pages, or none of the four topic words appears.

Removed from the denominator

Companies where no page body text could be read are removed from the denominator as not observable.

What we observed.

• Fully met — 4 of 21 companies: the FAQ covered cost, duration, the adoption process and support.
• Partially met — 3 companies: the FAQ covered only some of the four topics.
• Not met — 14 companies: no FAQ was found among the captured public pages, or none of the four topics was covered.

With this in place, the following become harder for a buyer to establish.

• a rough idea of cost and duration before making contact
• what needs to be prepared internally before adoption
• how far support extends after the contract starts

How to check your own site.

• Open your FAQ page and count how many of cost, duration, adoption process and support are answered.
• If you have no FAQ page, check whether the same content appears within your service pages.
• Read your last ten inbound enquiries and check whether the repeated questions are answered in the FAQ (needs internal data).
• Check that each answer is about contract, cost, duration or scope rather than a feature description.

If you leave it: You are left unable to confirm whether public information answers the questions that stall a purchase decision. This study does not assess whether an FAQ improves enquiry rates. What this study cannot conclude: Whether the FAQ actually removes purchase blockers.

Pricing information and route to a quote

Below the standard: 12 of 21 companies. "Below the standard" is the sum of partially met and not met. Counting only the companies that were clearly not met gives 9 of 21.

What we checked. Whether an actual figure appears in the captured page text, and if not, whether there is a stated route to a quotation. Whether the price is reasonable, and whether minimum terms or contract units are sufficient, were not assessed.

Fully met

An actual monetary figure appears anywhere in the captured page text — not necessarily as a price. Capital figures or amounts inside a case study also count.

Partially met

No figure appears, but there is a stated route to an enquiry or individual quotation.

Not met

Neither a figure nor any statement about pricing could be observed.

Removed from the denominator

Companies where no page body text could be read are removed from the denominator as not observable.

What we observed.

• Fully met — 9 of 21 companies: an actual monetary figure appeared on a public page.
• Partially met — 3 companies: no figure appeared, but there was a stated route to an enquiry or individual quotation.
• Not met — 9 companies: neither a figure nor any statement about pricing could be observed.

With this in place, the following become harder for a buyer to establish.

• whether the work might fit the available budget
• whether to keep the supplier on the shortlist
• whether an internal estimate can be shared

How to check your own site.

• Open your main pages and use in-page search (Ctrl+F / Cmd+F) for a currency unit, then check whether any actual figure appears.
• If you do not publish prices, look for a place that says so explicitly.
• Check whether the basis for a quotation — scope, minimum term, whether there is a setup fee — is stated.
• Run the same check on three competitors and see whether you publish noticeably less.

If you leave it: You are left unable to confirm whether a buyer can work out your terms from public information alone. What this study cannot conclude: Whether the price itself is reasonable.

Click depth

Below the standard: 12 of 21 companies. "Below the standard" is the sum of partially met and not met. Counting only the companies that were clearly not met gives 0 of 21.

What we checked. How many clicks it takes to reach each captured page by following internal links from the top page. The scope is the set of pages captured in this study; the actual reach of real visitors was not observed.

Fully met

Every captured page is reachable within two steps from the top page.

Partially met

Some pages are three or more steps deep, or not reachable from the links in the captured set.

Not met

Every page other than the top page is unreachable from the captured links.

Removed from the denominator

Companies whose top page could not be captured, or with fewer than two captured pages, are not assessed and do not appear in the totals.

What we observed.

• Fully met — 9 of 21 companies: every captured page was reachable within two steps from the top page.
• Partially met — 12 companies: some pages were three or more steps deep, or were not reachable from the links in the captured set.

With this in place, the following become harder for a buyer to establish.

• whether a visitor can get to your main service or pricing page without hunting
• whether someone arriving from search can move on to the page they need next
• whether an important page has fallen out of the navigation

How to check your own site.

• Start on your top page and count the clicks to your main service page.
• Do the same for pricing, case studies and contact.
• Check that all four are linked from the global navigation or the footer.
• Open a page that people commonly land on from search and check you can move on from there.

If you leave it: You are left unable to confirm whether a route to your important pages exists. Actual reach and drop-off were not observed. What this study cannot conclude: The actual rate at which visitors reach the page.

Form label implementation

Below the standard: 12 of 13 companies. "Below the standard" is the sum of partially met and not met. Counting only the companies that were clearly not met gives 12 of 13.

What we checked. Only whether a form field has a label element bound to it by id. Wrapping the input in a label, or using aria-label, was not counted as a label here. Input examples, explanations of why a field is required, error handling and pre-submit confirmation were not assessed in this study.

Fully met

On every page with a form, all visible fields carry a label bound by id.

Partially met

Some pages have fields without a label.

Not met

Every page with a form has fields without a label.

Removed from the denominator

Companies where no form appears in the server-returned HTML of any captured page are removed from the denominator as not observable. Forms injected by JavaScript are not visible to this method.

What we observed.

• Fully met — 1 of 21 companies: every visible field on every form carried a label.
• Not met — 12 companies: every page with a form had fields with no label.
• Not observable — 8 companies: no form was found on any captured page, so the company was removed from the denominator. These are removed from the denominator, not counted as failures.

With this in place, the following become harder for a buyer to establish.

• what to enter in each field
• what a field refers to once a placeholder disappears
• whether assistive technology can announce the field name

How to check your own site.

• Open your contact form and check that a field name is shown above or beside every input.
• Click into a field and type one character; confirm the field name is still visible after the placeholder disappears.
• Search the page source for "<input", then check that a "<label for=" exists with the same value as its id. A field with no id counts as unlabelled here even if a field name is visible.
• Open the same form on a phone and confirm the field names are not truncated.

If you leave it: You are left unable to confirm whether people can tell what each field is asking for. This study did not observe the actual experience of filling in the form, or abandonment. What this study cannot conclude: The actual experience of filling in the form.

04 / All approved results

"Below the standard" counts partially met plus not met. "Not met" is counted separately in its own column. The two are never merged, because the same check produces very different numbers depending on which one you use.

What the remaining checks actually looked at.

Company and operator details

Whether words for four items — address, representative, founding date and capital — appear in the company information page text, footer included. Whether an actual value follows each word, and whether the statements are accurate, were not assessed.

Breadcrumbs

Whether BreadcrumbList structured data is present on the page. Whether breadcrumbs are visible on screen was not part of this machine check.

Structured data

Whether structured data (JSON-LD or Microdata) is present in the server-returned HTML of each page, and if so which types. RDFa, and anything injected later by JavaScript such as through a tag manager, was not detected. Whether rich results appear in search was not assessed.

Canonical URL

Whether each page declares a canonical URL and whether that value matches the actual URL. Which URL a search engine actually selected was not assessed.

Search snippet basics

Whether each page has a title and a meta description, and whether the same title is reused across pages. What search engines actually display was not assessed, because they rewrite it.

Template consistency

Whether the share of pages with a single main element and the share with structured data are consistent across the site. Because this check only looks at variation, a site where neither is present on any page also counts as fully met. Presence itself is judged separately by the structured-data check. Whether any inconsistency is deliberate was not assessed.

Confidence intervals, variable definitions and the downloadable aggregate are on the data page. All 57 anonymised observation patterns, each traceable to the finding IDs behind it, are published on the Japanese page.

Aggregate data and variable definitions（日本語）

05 / A 30-minute check on your own site

Everything here can be done with a browser and your public pages. The standards are the same ones used in this study. For anything that applies to you, decide which of the four categories below it belongs to before starting work.

• Each of your five main pages has exactly one H1 and no skipped heading levels
• Article and case-study pages carry a date that corresponds to a real content update
• Every form field has a label element bound by id (placeholder text does not count)
• Your company information page states address, representative, founding date and capital
• Lower-level pages have both a visible breadcrumb and BreadcrumbList structured data
• Your FAQ answers cost, duration, the adoption process and support rather than describing features
• Even without published prices, the basis for a quotation and the route to one are stated
• Every main template carries structured data whose type matches the page
• Every page declares a canonical whose value matches the actual URL
• Important pages are reachable within two clicks of the top page (a page needing a third click counts as partially met)
• Main pages have a title and meta description, with no duplicates
• The main element and structured data are present consistently across template types

• A. Can be fixed with published information alone — wording or content you add.
• B. Needs a code or template change — you have to change what the component or CMS outputs.
• C. Verify after connecting internal data — connect GA4 and Search Console first, then measure.
• D. Needs a business decision — someone has to decide what may be published.

06 / What a change specification has to contain

"Improve the headings" or "add an FAQ" is a task name, not something anyone can implement or verify. Each of the 63 change specifications produced in this study states the target page type and surface, the observed facts behind it, the change itself, the completion criteria, how to verify it from public information, and what to verify once internal data is connected.

• Target — page type and the specific surface being changed.
• Observed facts — the audited statements the change answers to.
• Change — the steps to carry out.
• Completion criteria — what has to be true before it counts as done.
• Verification from public information — what to re-observe from outside.
• Verification after connecting internal data — what to measure in GA4 or Search Console.

Where a change touches code, the specification also fixes the component or template type, the props or CMS field, what must appear in the initial HTML rather than only after rendering, how it lines up with the structured data, where the canonical is generated, and the test that stops the problem coming back.

An approved change specification, published in full（日本語）Anonymised, with every field as it was written and reviewed.

07 / What to look at once analytics is connected

This study is not connected to GA4, Search Console or any CRM. That is exactly why the next measurement should be decided in advance, per finding.

Behaviour that is not recorded by default — form starts, abandonment, submissions — needs a measurement event defined first. This study did not observe those events, and we do not write as though they were already in place.

08 / Method and limits

• 21 Japanese B2B technology and professional-services companies. Candidates were gathered per sector and selected in a fixed-seed order. This is not a random sample and the proportions cannot be generalised to any industry.
• 8 to 12 pages captured per company.
• 75 checks across three layers and eight stages.
• Generative AI responses were observed through an API. They are not treated as equivalent to what the consumer ChatGPT or Perplexity products display.
• Public information only. No GA4, Search Console or CRM data was connected.
• AI assists with classification and drafting; conclusions are not finalised by AI alone.
• Anything that could not be observed is removed from the denominator and reported separately. It is never converted into a failure.
• Individual company results and any ranking are not published.
• A second review and a random re-review have not been carried out.

Dataset v1.0, methodology public-web-sales-audit-v1.0. The claims approved for citation in this dataset concern the structure of public websites. No claim about AI recommendation rates, citation rates or answer accuracy is included.

How the public-information audit works（日本語）

09 / Check the highest-priority break on your own site

This study publishes aggregate figures and anonymised patterns only. For your own site, the free diagnosis returns one change specification.

Public information only. No card registration. No automatic transition to a paid contract. Online application and card payment for the recurring diagnosis are still being set up; adoption is discussed through the existing enquiry route.