Organic clicks dropped by 14% while rankings stayed stable: a B2B SaaS case study on the impact of AI search

This is an anonymized case study about an international B2B SaaS company operating in a regulated market. I am not using the company’s name because the project was not public client work, but the situation itself will probably feel familiar to many B2B companies: a lot of content, strong subject-matter expertise, years of organic visibility, and still, declining organic search traffic.

More than 500 blog posts, but traffic was still declining

The website had more than 500 blog posts. At first, that looked like a strong foundation, and in many ways it was. Over the years, the company had built a large library of informational content around its market: regulatory topics, country-specific requirements, explainers, guides, and basic industry concepts. From a traditional B2B SEO perspective, this had been a logical strategy for a long time. If someone searched for a problem or a regulatory question, there was a good chance they would come across the company.
The problem became visible when search traffic started to weaken. It did not collapse from one month to the next. It declined gradually. Over a six-month period, clicks fell by 14%, which meant roughly a 2–3% decline per month. On its own, a monthly 2–3% drop can easily look like noise or normal fluctuation. But taken together, a 14% decline starts to suggest that something is changing.
What made the situation more interesting was that impressions and average position did not decline. In some query groups, there was even a slight improvement. This was an English-language website active across several major markets, bringing in roughly 6,000–7,000 organic clicks per month. So this was not a tiny sample where a few moving queries could distort the whole picture.
This is the kind of situation where it makes sense to carefully rule out the usual explanations. If rankings are not getting worse and impressions are not falling, but clicks are still declining, then it may not be a classic ranking problem. Of course, changes in titles or meta descriptions, either on your own site or on competitors’ sites, can affect click-through rate. In this case, however, we were able to control for that as well, because competitor pages were being monitored continuously, and we knew exactly where and when content changes had been made on the client’s own site.

Why was traffic declining?

After ruling out major changes on the client’s site and on competitor pages, the growth of AI search looked like a plausible explanation. Not as a single proven cause, but as a market shift that fit the data well. The pattern was not that the company had disappeared from search results. It was more that visibility was still there for many informational searches, but fewer people were clicking through. That distinction matters. When a page loses rankings, the story is usually simpler: it ranks lower and gets fewer clicks. Here, in many cases, the visibility remained, but the user no longer necessarily needed to click.
This was especially visible for topics with an explanatory intent. For example, when someone searched for a regulatory concept, a country-specific requirement, or a basic process. For these kinds of questions, an AI Overview, a Bing Copilot answer, or a ChatGPT-like search experience can often provide a good enough first answer. When that happens, the user has less immediate reason to continue to a blog post.
This does not mean that those pieces of content suddenly became worthless. That would be too strong a claim. It means their role changed. In the past, an informational blog post often acted as an entry point to the website. Today, the same content may work more as background material. It can help cover a topic, support entity building, appear as a source, or strengthen topical relevance, but it may not bring the same number of clicks it did a few years ago.
In practice, this means that for classic TOFU content, it is no longer enough to look only at impressions and average position. You also need to understand whether the user still needs to click at all. If the first answer is already available in the AI interface, the user will only continue if they need something deeper, more specific, or closer to a decision. This is where the focus of the project shifted. The main question was no longer how to win back every lost informational click. It was which pages could still play a meaningful business role in this new search environment.

How we tried to measure the impact of AI search

One of the main difficulties was that AI search traffic cannot be measured cleanly as a separate channel. In Google Search Console, you can see search queries, but queries that appeared in Google AI Overviews or AI Mode are not marked separately. So you cannot simply filter for “AI searches”. You have to approach the problem indirectly.
As a first step, I used regex in Search Console query data to filter search terms longer than eight words. I also looked separately at question-based searches. These do not prove that a query came from AI search, but they are useful starting points because AI-style searches are often longer, more conversational, and more specific than traditional Google searches. Some AI search analysts use 10+ word queries as a proxy. I used 8+ words as a slightly broader filter, not as proof, but as a practical signal worth investigating.
The next step was to review these queries with AI tools as well, to separate the ones that genuinely looked like AI-style search patterns. It is not a perfect method, but it is useful in practice, especially when you are not trying to draw conclusions from a single query, but looking for broader patterns.
I applied the same logic to Bing Webmaster Tools data. There, I looked at how many impressions these longer, question-like, or AI-style queries generated, how many clicks they brought, and whether any measurable conversions came from them.
This mattered because impressions alone do not say much. For a B2B SaaS company, the main question is not how often the brand appears for an informational search. The real question is whether those appearances generate meaningful traffic, and more importantly, visitors with business value.
The pattern was quite clear. The company had visibility for many informational or AI-style searches, but these did not produce proportional traffic or conversions. This supported the assumption that the role of classic TOFU content is changing. In many cases, users already get their first answer inside the AI interface, and they only click through when they need something deeper, more specific, or closer to a purchase decision.

What the Bing data showed

One data point showed the scale of the gap very clearly. During the period I reviewed, Bing Webmaster Tools showed more than 100,000 impressions in Copilot and Bing AI Answer surfaces. The traffic from these appearances amounted to 20 users on the website. That is roughly a 0.02% click-through rate. It is hard to find a cleaner example of the gap that has opened between being visible in AI surfaces and actually getting a website visit.

The value of informational searches has changed

The company appeared for many informational queries where users would previously have been more likely to click through. These were typically definition-based, explanatory, regulatory, or basic “how does this work?” searches. In other words, exactly the kind of topics where an AI answer can now often provide a good enough first response. Not always a perfect answer, not always a deep enough answer, but enough for many users.
This distinction is important. TOFU content has not suddenly become worthless. Its role has changed. If someone only wants to understand a basic concept, an AI Overview, a Bing Copilot answer, or a ChatGPT summary may satisfy the need. In that case, the click may never happen. The user may never reach the website, even if the website’s content helped the AI system understand the topic.

We did not start by rewriting content. We started by looking at the data.

The first step was not to choose twenty blog posts and start updating them. Before changing content, we needed to understand what kind of traffic was declining, which pages were affected, and where the remaining traffic was still creating business value. That required bringing several data sources together.
Search Console and Bing Webmaster Tools showed the visibility side: which queries the company appeared for, where impressions remained stable, where clicks declined, and which topics showed the biggest changes in click behavior. In BigQuery, I looked at traffic and conversion data, with a separate focus on AI-search-related referral sources, conversions, and the role different page types played in the user journey.
AI search traffic still cannot be measured perfectly. That is important to say clearly. Referral data only shows the visible part of the story. If someone discovers the company in an AI answer and later returns through a brand search or direct traffic, analytics may no longer attribute that visit to AI search. But even with this limitation, useful patterns appear. You can see which topics generate a lot of visibility but little conversion, and which lower-traffic pages bring visitors with higher business value.

Content volume was not the missing piece

The most important finding was that content volume was not the problem. In fact, the site was probably too heavily weighted toward informational content. The hundreds of blog posts gave the company strong topical coverage, but the search environment had changed. Articles that once worked as entry points now often play more of a supporting role. They can still help cover the topic and strengthen entity signals, but they may not generate the same clicks or the same business value as before.
That is why the project shifted toward improving MOFU and BOFU pages. The goal was not to win back every lost TOFU click. In many cases, that would be unrealistic. Instead, we needed to identify the search situations where users were no longer just trying to learn, but were starting to make a decision. That might mean a solution page, an industry landing page, a use case page, a comparison page, or a page that answers a specific business problem.

Why MOFU and BOFU pages became more important

In B2B SaaS, these pages often do not bring the highest traffic. But that is not their job. A good MOFU or BOFU page is not built for volume. It is built for relevant decision moments. A finance, compliance, or operations leader is unlikely to contact a vendor because of a basic definition. They are more likely to do so when they already understand the problem, see the risk, and want to know what options they have.
The content changes were based on query fan-out analysis. Query fan-out means that one search does not remain one question. It is broken into several related sub-questions. In classic Google search, we are used to the user typing in a keyword or question and the search engine returning results. In an AI search environment, this often works differently. When someone asks a more complex question, the AI system does not necessarily try to answer it as a single unit. It breaks the question into related parts and looks for information, context, examples, or sources for each of them. That breakdown is the fan-out.

Query fan-out helped reveal the real decision questions

In a regulated B2B market, user questions rarely stop at “what is this?”. In a real decision process, the questions become more specific. Which countries does this affect? What is the risk of getting it wrong? Which internal teams need to be involved? When is a manual process enough? When is software needed? What integrations are required? How should different providers be compared?
These are no longer simple informational questions. They are decision questions. And because of that, they require a different type of content.

The content did not just become longer. It got a different job.

In practice, this did not simply mean writing longer pages. Often, length was not the point at all. The real goal was to make the page more useful for decision-making. The content started to include more concrete decision criteria, common mistakes, internal objections, implementation questions, compliance risks, and sections that explained when the solution is needed and when it is not. That last part is especially important. A good B2B page does not try to convince every visitor immediately. It is often more credible when it helps the reader understand the problem more clearly.

E-E-A-T is not decoration in a regulated market. It is a trust layer.

Another important part of the project was strengthening E-E-A-T signals. In a regulated market, this is not decoration. It is trust infrastructure. If a company communicates about topics with legal, financial, tax, or compliance implications, it is not enough for the content to be well written. It must also be clear why the company is credible, who stands behind the expertise, what sources the content relies on, and how the company is connected to the topic within the market.
Several elements were worth improving: author pages, expert references, external mentions, structured data, update dates, company profiles, and entity signals. Connecting authors as entities to relevant education, experience, and publications was an important step. For one author, we even managed to create a Wikidata page, which can be a useful signal for AI systems.

Not every page type should be handled the same way

Another lesson from the project was that page types need to be handled separately. A blog post, a solution page, an industry landing page, and a comparison page do not have the same job. They may need different schema, different internal linking, different CTAs, and they play different roles in the user journey. Many SEO projects blur these differences, but in practice they matter a lot.

Results had to be measured beyond organic traffic

For that reason, I did not measure the project only by looking at whether organic traffic increased. In an AI search environment, that is not enough. It was more important to understand whether visibility improved for decision-stage topics, whether AI/referral traffic appeared, how those visitors behaved, whether they converted, and whether MOFU and BOFO pages started to play a larger role in conversion paths.
I measured the results over a six-month period. The first month served as the baseline, and I compared it with the sixth month. This is not a perfect measurement model, but for this kind of project it is long enough to avoid drawing conclusions from weekly fluctuations.

Decision-stage visibility improved

Visibility for decision-stage, AI-style searches improved meaningfully. Based on filtered query data from Bing Webmaster Tools and Search Console, impressions for these searches increased by 42.1% compared with the position six months earlier. In numbers, this meant an increase from 952 estimated impressions to 1,351 impressions.
This should not be overvalued on its own. An impression is not a business result. But it did show that the revised content and decision-stage pages were appearing in more relevant search situations.

AI/referral traffic became measurable

During the one-month period reviewed, AI/referral sources brought 423 visits, which generated 30 conversions. This should not be framed simply as “a lot” or “a little”. It is better to treat it as a baseline. By conversions, I mean marketing conversions such as book-a-call actions and newsletter sign-ups. The more important signal was that this traffic segment became measurable, produced conversions, and continued to grow in the following months. The company had roughly 500 conversions per month overall.
To me, this was a stronger signal than a simple increase in organic sessions. The goal of the project was not to bring back every lost informational click. It was to help the company show up better in the search situations where the user is closer to a business decision.

The main lesson: not every lost click is worth winning back

For me, the main lesson from this case study is that in B2B SEO, content volume is becoming less valuable on its own. Hundreds of blog posts can be a strong foundation, but they do not protect a company from the fact that some informational searches are moving into AI-generated answers. The role of the website does not disappear, but it does change.
In a situation like this, the best question is not “how do we win back every lost click?”. A better question is: which pages need to support real buying decisions, and how can we improve them so they perform better in search, in AI search, and from a business perspective?