How do I know whether our website is actually working?

Every month a report arrives. Sessions are up, bounce rate is down, and a chart shows a line going roughly the right way. Meanwhile your sales team says new clients almost never mention the website, and the contact form brings in mostly suppliers trying to sell you something. You are left wondering whether the site does anything at all.

A B2B website is working if it brings in enquiries from the right people, helps buyers who already know you decide in your favour, and gives search engines and AI assistants an accurate picture of what you do. Traffic on its own proves none of those.

The last job matters more each year. In G2's April 2026 survey of 1,076 B2B software buyers, 51% said they now start their research in an AI chatbot, up from 29% a year earlier. G2's sample is software buyers, but we see the same pattern in client work: at Helm Club, "1 in 5 new members this year came through an AI recommendation." This guide shows you how to judge your own site in an afternoon, without specialist tools.

#What does "working" mean for a B2B website?

It means four jobs done well: bringing in qualified enquiries at a steady rate, helping buyers who already know you decide in your favour, giving Google and AI assistants an accurate picture of what you do and for whom, and being somewhere your team is proud to send people.

Traffic, page views and bounce rate help you diagnose problems with those jobs, but they are not the jobs. A site with modest traffic that brings in two good enquiries a month beats a busy one that brings in none. To put a financial value on the site, see how to measure the return you get from your website.

#How to evaluate your website, step by step

#Step 1: Count the enquiries that came through the site

Start with the result. List every enquiry from the last three to six months that came through the website: forms, calls to the number on the site, emails to the address listed, meeting bookings. Note which were the right kind of client and which became work. Then ask sales how many clients won in that period looked at the site first, and what they said.

From now on, ask every new enquiry how they found you, with "ChatGPT or another AI assistant" among the options, and record it in your CRM. We cover combining those answers with analytics in how to tell which new leads came from ChatGPT.

#Step 2: Check where visitors come from, including AI assistants

Look for visits from chatgpt.com, claude.ai, perplexity.ai, gemini.google.com and copilot.microsoft.com. Three details trip people up.

• GA4's AI channel has gaps. GA4 gained a default "AI Assistant" channel on 13 May 2026, and Google's channel definitions say it covers ChatGPT, Gemini, DeepSeek, Copilot and Grok, while clicks from AI Overviews and AI Mode count as Organic Search. Claude and Perplexity are not on the list, so their visits sit in Referral.
• ChatGPT tags its links. OpenAI says ChatGPT "automatically includes the UTM parameter utm_source=chatgpt.com in referral URLs", so filter on that source too.
• App visits often show as Direct. Links tapped in the ChatGPT app often lose their referrer on the way to the phone's browser (Nice Looking Data, July 2026). Direct visits to a deep service page nobody would type from memory are often unlabelled AI traffic.

To see the AI visits you can identify in one line, go to Admin, then Data display, then Channel groups in GA4. Copy the default group, add a channel called "AI assistants" with the condition "Session source matches regex", and paste chatgpt\.com|chat\.openai\.com|claude\.ai|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com. Move it above Referral and save. Treat the result as a minimum. For Google's own AI features, the generative AI reports Search Console rolled out to all sites by 31 August 2026 show how often your pages appear in AI Overviews and AI Mode.

#Step 3: Read the site as a buyer would

This is the most useful step and the one most firms skip. Open your homepage on a phone and give it ten seconds. Can you tell who the firm helps, what it does and where? Then check your main service page: does it describe the problem in the buyer's words, show proof such as case studies with outcomes, show who you would work with, and make the next step obvious?

If a new salesperson could not explain your firm from those two pages, neither can a buyer or an AI assistant. We cover this in what makes a buyer trust your website enough to get in touch.

#Step 4: Follow the path to an enquiry

From each of your five most visited pages, count the clicks to a way of getting in touch. Then fill in your own contact form on your phone. Does it work, and how quickly does anyone reply? In our experience this step turns up broken forms and unchecked inboxes far more often than founders expect, and none of it shows in a traffic report.

#Step 5: Ask AI assistants what they know about you

Write five questions a buyer would ask, such as "best IT support for accountancy practices in the North West". Ask each in ChatGPT and Google's AI Mode first, then Copilot and Gemini, then Perplexity and Claude, in a fresh chat each time. Record who is named and how you are described.

Then ask each assistant what your firm does. A vague or wrong description means your site is not stating the facts clearly, or other sources contradict it. A correct description with no recommendation means the assistants know you and have no reason to name you. That was ID-Pal's position: "Across 36 recorded AI answers to the questions ID-Pal's buyers actually ask, ID-Pal was named three times, and only for its brand or its Salesforce integration. AI knew who it was, and had no reason to recommend it."

Answers vary between runs, so for tracking keep a fixed list of 20 to 50 buyer questions, ask each several times a month in a clean session, and count a win only when it repeats, as set out in finding out whether ChatGPT recommends your firm.

#Step 6: Check the technical basics

Technical problems rarely win work but can stop a site working. Google's Core Web Vitals guidance sets three targets: main content within 2.5 seconds, a response to a tap or click in under 200 milliseconds, and a layout shift score under 0.1. Also check the site on a real phone, confirm https with no warnings, fix "not found" pages, and ask your developer to confirm that neither robots.txt nor a Cloudflare or firewall setting blocks the crawlers used by ChatGPT, Claude and Perplexity. The full list is in whether your website is technically ready for AI search.

#Step 7: Compare yourself with three competitors

Look at three competitors you meet in pitches the same way, and compare who gets named in Step 5. Look for the gaps a buyer would notice with both sites open side by side.

#How do you score the result?

Score the site from 1 to 5 on six questions, with one sentence of evidence beside each. The questions follow the lenses of our free Growth Review, which scores a firm out of ten on five: Found, Named by AI, Website, Authority and Pipeline. Here Website is split into three, because it is the part you control, Enquiries comes first as the result, and Authority and Pipeline, which sit outside the site, are left out.

The maximum is 30, and the bands are a rule of thumb. 24 or more is an average of at least 4, mostly "yes": the site is broadly working. 15 to 23 means it works in parts; start with the two lowest scores. Below 15 is an average under 2.5, more "no" than "yes": the site is probably costing you work. Whatever the total, a 1 or 2 on Enquiries outweighs the rest. Repeat the score quarterly, with the same person scoring.

#A worked example

Example: a fictional 40-person engineering consultancy. The firm and the figures are illustrative.

It chose two fixes from the lowest scores. First, rewrite the homepage and two main service pages to name the sectors served, each with a recent case study that leads with its result. Second, correct the old service list everywhere it appears, on the site, LinkedIn and directories, so AI assistants have one consistent description to draw on.

#Why does our website get traffic but no enquiries?

Usually for one of four reasons: the visitors are not buyers, the pages they land on do not speak to them, the route to contact is broken, or enquiries arrive and are not credited to the site. Check them in that order.

Falling numbers need a cause before a redesign. At ID-Pal, "Traffic to the pages that bring in enquiries had fallen by around 40 per cent since the NorthRow move, and nobody could say why." The cause was one no traffic report could show: "The redirect from northrow.com was technically correct, but nothing on either site told Google or the AI engines that NorthRow and ID-Pal were the same company."

#When should you fix the site, and when should you rebuild it?

Fix it if the problems are about content and clarity. Rebuild only if the platform or structure stops you fixing them.

Rewriting service pages, adding case studies and simplifying forms can be done on the existing site, faster than a rebuild and without risking your search history. Consider a rebuild when your team cannot edit pages without an agency, the platform is unsupported, or the structure no longer matches how you sell. We compare the options in template or custom website: which does a B2B firm need?, and explain why a site left alone drifts in whether to redesign or keep improving.

#What to do this month

1. List last quarter's website enquiries. Note which were qualified and which became work.
2. Add the AI channel to GA4. Use the pattern in Step 2, then compare it with Direct visits to your service pages.
3. Read your homepage and main service page on your phone. Write down what a stranger would understand in ten seconds.
4. Ask the main AI assistants five buyer questions. Record who is named and how your firm is described.
5. Test your own contact form on a phone, and time the reply.
6. Score the site on the six questions. Fix one thing in each of the two lowest before next month.

#Frequently asked questions

#How many enquiries a month should a B2B website generate?

There is no reliable benchmark, because client values and markets vary too much. Work backwards instead. For example, if you want 12 new clients a year, win one proposal in three, and half of qualified enquiries reach a proposal, you need about 72 qualified enquiries a year from all sources. If the website should supply a third, that is about two a month.

#How do I set up a channel for ChatGPT traffic in GA4?

The default AI Assistant channel misses Claude and Perplexity. Add a custom channel under Admin, Data display, Channel groups, matching session source against the regex in Step 2, placed above Referral. Many AI app visits still arrive as Direct, so pair it with a "how did you find us" question.

#What should a free website audit include?

It should start from your enquiries and your buyers' questions, not only an automated technical score. A useful audit covers what the site brings in, how it reads to a buyer on a phone, what Google and AI assistants say, whether the contact route works, the technical basics and two or three competitors. Our free Growth Review does this in 30 minutes: five scores, the buyer questions where you are not named, and three fixes in writing.

#How do I know if my website is putting buyers off?

Listen to the sales process. Prospects who ask questions the site should answer, who did not realise you offer something you have done for years, or who need a PDF after every call are telling you the site is not doing its job. In analytics, look for visitors who reach a service page and leave without clicking anything. The quickest test is to ask five recent clients what they thought when they first looked.

#Our traffic is falling but enquiries are steady. Should we worry?

Not necessarily. Many buyers now get answers from Google's AI Overviews or ChatGPT without clicking. If enquiries hold, the site may be doing its job with fewer, better-informed visitors. Check what AI assistants say about you and whether AI visits, including unlabelled Direct ones, are growing.

#How often should we evaluate our website?

Run the full score once a quarter and check enquiries and AI answers monthly. Assistants that search the web live can reflect a changed page within weeks, while answers drawn from a model's training change only when it is retrained, which takes months.

Most B2B websites that are not working have a handful of fixable problems rather than a fatal flaw. If you would like a second opinion, our free Growth Review scores your firm on the same lenses, shows the buyer questions where you are not named, and sends the three fixes we would make first, in writing.

#Sources