Where public attention comes from when search sends fewer clicks

Where public attention comes from when search sends fewer clicks

Between 21:19 UTC on 24 September and 06:39 UTC on 25 September 2026, our edge proxy served 77,221 page requests for trustlist.uk. We counted who asked for them. AI companies' crawlers and fetchers made 27,591 of those requests, or 35.7%. OpenAI's GPTBot alone made 15,285, and OAI-SearchBot, the crawler OpenAI uses for ChatGPT's search features, made another 5,359. Googlebot made 221, or 0.3%. In that window GPTBot fetched about 69 pages for every one Googlebot fetched. Meta's crawlers made 4,519 requests, Applebot 1,747, Bingbot 2,180.

The people side of the same log looks very different. Of 35,401 page views from browser-like user agents, 76.1% carried no referrer at all, 23.8% came from our own pages, and 27 came from a Google search page. None came from a social network. This is one nine-hour overnight window, a sample and not a month, and our analytics show that a good share of the browser traffic is an automated browser farm rather than people. But it matches what our longer data says. Across the 92 days to 18 September, Google Search Console recorded 652,463 impressions and 657 clicks for the site. Pages about one named business or product earned 528 of those clicks, about 80%, from roughly 5% of the impressions. Our broad ranking pages collected 255,000 impressions and 67 clicks.

Two things follow from those counts, and the rest of this article builds on them. First, machines now read the web far more often than search engines send people to it, and some of those machines answer people's questions directly. Second, when a person does arrive, they usually arrive because they were already looking for a specific, named business. Attention is going to whoever is named, cited and easy to verify.

This piece is part of a series. For what this meant for our own traffic, see what we learned trying to earn traffic for a directory. For whether search itself is fading, see is SEO dying or changing and the changing role of search engines.

What the wider evidence says about search clicks

Our numbers are small and specific to one site. The broader evidence points the same way, with some important limits.

The Reuters Institute's Digital News Report 2026, published on 16 June 2026 and based on surveys of about 2,000 people in each of 48 markets, reports aggregate data from the analytics company Chartbeat: Google traffic from organic search to more than 2,500 news sites fell by a third (33%) globally between November 2024 and November 2025, and by 38% in the United States. Those are news publishers, not B2B firms, and news is the category most exposed to AI summaries. But it is the largest measured sample we have found that was published with its source.

The same report found weekly use of AI chatbots such as ChatGPT, Perplexity and Google Gemini for news rose from 7% to 10% of people across its markets in a year. That is growth, but from a low base: nine in ten people still did not use a chatbot for news in a given week. It also found that 42% of people who use chatbots for news say they always or often click through from the answer to the original source. A citation in an AI answer can still send a visit. It just sends fewer of them than a list of ten blue links did.

The honest reading is that search has not gone away. People still search, and Bing and Google still crawl. What has changed is that a growing share of questions is answered before anyone clicks, and the answer draws on sources the asker never visits. Being one of those sources is now a form of attention in itself, even when it produces no visit.

Answer engines: how they choose what to cite

An answer engine is any tool that replies to a question in sentences and names its sources: ChatGPT with search, Perplexity, Microsoft Copilot, Google's AI answers, Meta AI. Each has its own rules for what it reads, and the rules are published.

OpenAI runs three separate agents, and its own documentation explains each. OAI-SearchBot is used to surface websites in ChatGPT's search features. GPTBot crawls content that may be used to train OpenAI's models. ChatGPT-User visits a page when a person's question needs it, and OpenAI says it is not used for automatic crawling. The practical point is in OpenAI's instructions to publishers: to appear in ChatGPT search results, allow OAI-SearchBot in your robots.txt file and allow requests from OpenAI's published IP ranges. Blocking GPTBot (training) does not remove a site from search; blocking OAI-SearchBot does. A firewall rule or a bot-protection default can block all three without anyone deciding to.

We let all of them in, deliberately. In our overnight sample ChatGPT-User made only 66 requests, which is the number of times a live ChatGPT conversation needed one of our pages in that window. That is the figure closest to real demand, and it is small. The 5,359 OAI-SearchBot requests are the preparation: an index being built so that our pages can be cited later.

Microsoft now tells site owners how often they are cited. Bing Webmaster Tools has an AI Performance report, and on 16 June 2026 Microsoft added four views to it in preview: Intents, Topics, Citation Share and Compare. Citation Share is, in Microsoft's words, the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query. For the first time a business can see which of its pages Copilot and Bing's AI answers rely on, and for which questions. If you do one thing after reading this article, verify your site in Bing Webmaster Tools and look at that report. Bingbot made 2,180 requests to our site in the overnight window, ten times Googlebot's count.

Perplexity has gone further and pays publishers. In August 2025 Jessica Chan, its head of publisher partnerships, told Digiday that subscription revenue from its Pro, Max and Comet Plus tiers goes into a pool, with 80% shared among publishers in its programme and 20% kept by Perplexity. Payment is triggered by direct visits through its Comet browser, by citations in answers and by use of the content in agent tasks. The programme is aimed at publishers, not at most B2B firms, but it shows where the model is heading: a citation is becoming a unit of value, whether or not a click follows.

What do these systems cite? None of them publishes a formula, and anyone who sells you one is guessing. What their documentation and our logs do show is simpler. They can only cite pages they are allowed to read, and a live fetcher such as ChatGPT-User goes to a page only when a question needs it. Our own evidence is that the pages people click are about a named business, with a specific answer on the page. We expect the same pages to be the ones worth citing, and we will be able to check that as the Bing report fills in.

Newsletters, communities, video and podcasts

Search and answer engines are the machine-mediated routes. The other routes are people-mediated, and they are growing for the same reason: people want a source they recognise.

Video. The Reuters Institute found that, averaged across 48 markets, social media and video networks are for the first time the single most widely used way of reaching online news, used by 54% of respondents. For a B2B firm the relevant platform is usually YouTube. Edison Research's Infinite Dial 2026, published on 12 March 2026, found that 84% of Americans aged 12 and over use YouTube monthly and 78% weekly. Those are US figures. They are useful because YouTube is also a search engine, and a short video that answers one customer question keeps being found long after it is posted.

Podcasts. The same Infinite Dial study found that 58% of Americans aged 12 and over had consumed a podcast in the last month, 45% in the last week, and 57% had both listened to and watched a podcast. Podcasts have become a video format as much as an audio one. For a small B2B firm, being a guest on an established show in your niche is usually worth more than starting your own: the audience already exists, and the episode page links back.

Creators and named people. The Reuters Institute found that 27% of people consume news from news-focused creators, and 46% from any kind of creator. The pattern, which we see in our own logs as well, is that attention attaches to a named source. People follow a person or a firm they know, rather than a platform's general feed.

Newsletters. We have not found a primary, recent, sourced figure on B2B newsletter readership that we are willing to print, so we will not give one. The argument for a newsletter does not rest on a statistic. An email list is the one audience channel you own. No algorithm decides whether your subscribers see your message, and no crawler takes the content without sending the visit. Our log's 76.1% referrer-less browser traffic is a reminder that a lot of real visits arrive with no trace of where they came from: typed addresses, bookmarks, links in email and messaging apps. Newsletter traffic tends to look like this.

Communities. Reddit's results for the second quarter of 2026, filed with the SEC on 30 July 2026, reported 130.3 million daily active uniques, up 18% on a year earlier, and 514.6 million weekly active uniques, up 24%. Its chief executive, Steve Huffman, put the reason in one line: in an increasingly automated web, the value of real human perspective has never been higher. For a B2B firm, the lesson is not to post adverts in forums. It is that buyers ask each other about suppliers in public, and those threads are read by people and by AI systems alike. Being mentioned accurately in them, by customers, is worth more than anything you post there yourself.

Marketplaces, launch sites, partner directories and events

Some of the most dependable sources of attention for a B2B firm are listings on other people's platforms, where the platform already has the buyer.

Software marketplaces. If you sell software that runs on or next to a large platform, its marketplace is a sales channel as well as a listing. Microsoft's own documentation says it charges a 3% standard store service fee on offers bought through Microsoft Marketplace, and supports flat-rate, per-user and metered pricing. Buyers often prefer this route because the purchase can go on an existing Microsoft bill. AWS, Google Cloud, Salesforce, Atlassian and HubSpot also run marketplaces, each with its own terms and fees, which you should read before listing. We cover how this affects pricing in how software products make money now.

Integration and partner directories. Every serious SaaS product lists the tools it connects to and the partners who implement it. Those directory pages rank for the product's own name, which means they appear when a customer is already looking. For an IT services firm, being a listed partner of the software your clients use is often the most direct route to them.

Launch sites. Our own launch board shows how crowded this has become. We counted 60 launches dated September 2026: 28 AI models and 32 products. Many of the new products we reviewed were relaunches or had no named operating company, and several published no price. A launch-day spike is real attention, but it is short, and it rewards products that already have an audience. If you launch, make sure the page a visitor lands on names the company behind the product and states a price. Those are the first two things a buyer checks and the first two things many launches leave out.

Directories. We are one, so treat this paragraph with the scepticism it deserves. Our Search Console figures say that listing pages, one per business, earned 80% of our clicks. A directory listing is useful to a firm only if it is complete and accurate. On 24 September we drew 1,000 of our company listings at random and found that 171 websites no longer led to the business as listed. A quarter of our live listings carry fewer than 50 words of profile text. We are rewriting those from each firm's own site, but the firm itself is the best source of its own facts. Check every directory profile you have, at least once a year.

Events. Trade shows, meet-ups and conference talks remain one of the few routes where the buyer, the seller and the conversation are all in one place. They are expensive in time. Their lasting value for attention is the material they produce: a recorded talk, a slide deck, a write-up. Each of those gives people and machines something to find and cite afterwards.

Why the named-business page matters more than it used to

Our 92-day figures show two kinds of page behaving very differently. Pages about one named business got 1.6% click-through at average positions of 6 to 9. Ranking pages that list many firms got 0.03% at positions 60 to 80. Some of the difference is position. Much of it is intent. A person who types a firm's name, or a product's name and "pricing", knows what they want. A person who types a broad category phrase is often satisfied by the answer at the top of the page.

Answer engines strengthen this pattern. A broad question, such as which accounting software suits a small firm, is exactly what an AI answer handles without a click. A specific question, such as whether a named supplier offers a service in Leeds, or what a product costs after its latest price change, needs a source, and the source that gets cited is the page that answers it plainly.

So the work that earns attention now looks less like publishing more pages and more like making the pages about you complete, current and consistent everywhere they appear: your own site, your marketplace and partner listings, your directory profiles, your social profiles. If an answer engine finds three different addresses for your business, it has no reason to trust any of them.

A practical attention plan for a small B2B team

This plan assumes a team of one to five people with a few hours a week to give to it. It is ordered by what we think pays back first.

Week one: make sure you can be read

1. Check your robots.txt and firewall. Confirm that Bingbot, Googlebot and OAI-SearchBot are allowed. Decide separately whether to allow training crawlers such as GPTBot; that is a policy choice, not a visibility requirement. Ask whoever runs your hosting or security service whether a bot-protection setting is blocking crawlers you want to let in.
2. Verify your site in Bing Webmaster Tools and Google Search Console. Look at the Bing AI Performance report for which of your pages are cited, and at Search Console for which queries already bring you clicks.
3. Write down your facts in one place. Legal name, trading name, address, services, sectors, prices or price ranges, founding year, team size. This becomes the source for every profile.

Month one: fix the pages about you

1. Make one page per service or product that answers the questions a buyer asks before calling: what it is, who it is for, where you deliver it, what it costs or how it is priced, and how to start.
2. Put a date on anything that changes. Prices, terms and certifications go out of date. A dated statement can be trusted and cited; an undated one cannot.
3. Update every external profile from your fact sheet: marketplaces, partner directories, business directories including ours, LinkedIn, Google's business profile. Remove duplicates you no longer control where you can.

Each quarter: produce things worth citing

1. Publish one piece of original evidence. Something you counted or measured: response times across your support tickets, the most common causes of failed projects you were called in to rescue, what your clients' licence costs did this year. Original figures are the scarcest thing on the web, and they are what journalists, newsletters and answer engines quote.
2. Turn it into three formats. A page on your site, a short video, and a newsletter issue. Offer it to one podcast or event in your niche.
3. Ask satisfied clients to say so where buyers look, in their own words, on the platforms they already use. Never script reviews, never pay for them, and never post them yourself.

What to stop doing

• Stop measuring success only by Google clicks. Track Bing citations, direct and referrer-less visits, newsletter sign-ups and enquiries that mention where the buyer heard of you.
• Stop producing broad articles that restate what an AI answer already says. They collect impressions and no visits.
• Stop spreading effort across every social platform. Read will social media survive the AI change before deciding where yours goes.

The short version: be readable by the machines that now answer your buyers' questions, be accurate everywhere you are named, and produce a small amount of evidence that nobody else has. Attention follows the named, the specific and the verifiable.

Sources

• TrustList edge-proxy logs, 24 September 21:19 to 25 September 06:39 UTC: page requests by user agent and browser referrers. TrustList, counted 25 September 2026. One overnight sample, not a month.
• TrustList Google Search Console export, 92 days to 18 September 2026: impressions, clicks and click-through by page type. TrustList, 18 September 2026.
• Digital News Report 2026: executive summary: weekly AI chatbot news use (7% to 10%), 42% click-through, Chartbeat organic search decline (33% globally, 38% US), 54% social and video, creator figures, 48 markets. Reuters Institute for the Study of Journalism, 16 June 2026.
• Overview of OpenAI crawlers: what OAI-SearchBot, GPTBot and ChatGPT-User do and how to appear in ChatGPT search. OpenAI (read 25 September 2026).
• New AI visibility insights in Bing Webmaster Tools: Intents, Topics, Citation Share and Compare. Microsoft Bing, 16 June 2026.
• How Perplexity's new revenue model works, according to its head of publisher partnerships: 80/20 revenue pool and the three payment triggers. Digiday, 26 August 2025.
• The Infinite Dial 2026: US podcast and YouTube use, ages 12+. Edison Research at SSRS, 12 March 2026.
• Reddit reports second quarter 2026 results: daily and weekly active uniques and the chief executive's comment. Reddit, Inc. (Form 8-K exhibit), 30 July 2026.
• Microsoft Marketplace transact capabilities: 3% standard store service fee and supported pricing models. Microsoft Learn, 22 July 2026.
• TrustList launch board: 60 launches dated September 2026 (28 AI models, 32 products). TrustList, counted September 2026.
• TrustList catalogue checks: 171 of 1,000 random company websites no longer lead to the business (24 September 2026); about 24,350 live listings under 50 words of profile text (25 September 2026). TrustList.