I analyzed 10,172 blog posts from 254 B2B SaaS companies, dated November 2025 to October 2026.
I wanted to understand what B2B SaaS companies are publishing, how they're structuring content, and whether content is becoming too similar.
I also looked at how AI is influencing content and how companies promote their products.
Here's what I found.
- FAQs increased from 17.3% to 37.8%.
- Articles with tables increased from 16.8% to 35.6%.
- Median article length rose from 1,390 to 1,760 words.
- 30.2% of articles had no editorial image, video, or table.
- 27% of eligible “What is X?” articles followed the same four-part structure.
- Em-dash use fell from 66.7% to 36.6%.
- Stock marketing phrases fell by 63%.
- 18% of detected numerical claims had a clearly attributed external source.
How B2B SaaS articles are formatted
FAQs increased from 17% to 38%
In November 2025, 17.3% of articles included FAQs. In October 2026, 37.8% did.
Tables followed, rising from 16.8% to 35.6%. Both increases remained substantial when comparing posts from the same publishers.
These elements often appear together. Of articles with FAQs, 34% also included a summary block. Only 12% of articles without FAQs had one.
This trend extended to tables, too. Around 40% of the 223 companies compared increased their use of both FAQs and tables during the year.
The likely reason FAQs increased was the advice around generative engine optimization and getting cited by AI, which also coincided with Google's decision to stop displaying FAQ rich results in May 2026.
GEO advice continues to recommend FAQs as a way to make answers easier for AI systems to find and use.
There's some external research supporting that advice. In July 2026, seoClarity reported that adding FAQ sections increased AI citations in a client test tracking around 1,000 prompts. When the FAQs were removed, citations fell again.
A 2026 study by SISTRIX looked at 96,418 URLs and found that pages using FAQPage markup were cited 1.5 times as often by ChatGPT, Google's AI Mode and AI Overviews as comparable pages from the same websites.
The technical implementation also varied. Only 42% of articles with a visible FAQ section included FAQPage schema, the code that identifies questions and answers to search engines. Around 5% had the markup without a detectable FAQ section.
The increase wasn't universal. Twenty-one companies reduced their use of FAQs, 16 reduced tables, and 35 reduced summaries. In Marketing & CRM, summary use fell from 29% to 23%.
Summary sections became more common overall, although short TL;DR blocks barely changed. Tables appeared in 26.5% of all articles.
89% of key takeaways appear near the top of articles
Around 18% of articles included a summary, TL;DR or key-takeaways section. Of those with key takeaways, 89% placed them near the beginning, usually with four bullet points highlighting the most important information.
Another 45.5% included a table of contents, giving readers a way to jump directly to the sections they needed.
This follows a wider trend in content formatting. GEO advice recommends placing important information near the beginning of an article, using clear headings, and making answers easy for AI search engines to identify.
Additionally, for 2,226 articles with a question in the primary title, 40.2% answered that question in the opening sentence.
These approaches all prioritize making individual parts of an article understandable without requiring someone to read it from beginning to end. These are common tactics when optimizing for AI search.
65% of question headings lead directly to an answer
In the 10,172 articles analyzed, sections contained a median of 91 words. Articles also had around six H2 subheadings each.
Question-style headings appeared in roughly 42% of articles, a figure that remained relatively stable during the year. But 65% of question headings outside FAQ sections gave a direct answer in the first sentence beneath them.
This relates to how search and AI systems break articles into smaller passages when looking for information (chunking). Clear headings and self-contained explanations can make passages easier to find. But making every section extremely short doesn't necessarily help.
Interestingly, articles weren't moving in this direction.
The percentage of articles containing three consecutive H3 subsections, each under 50 words, fell from 21% to 14%. Very short sections containing fewer than 40 words and no list or table also became less common, dropping from 14.1% to 10.8%.
The typical section actually grew from 87 to 96 words between the first and second halves of the year. H3 usage declined, and articles contained approximately 1.2 fewer headings per 1,000 words.
These findings connect to my separate content chunking experiment.
When the same content was presented with its original headings and paragraphs intact, the retrieval system found the correct passage in its top five results 80% of the time. That fell to 68% when the content was presented as dense prose.
Breaking the same information into short, automatically generated Q&A sections performed much worse, with just 12% correct retrieval.
The experiment suggests that effective chunking depends on how information is organized, not simply how short each section is.
Articles over 2,500 words increased from 20% to 26%
Long-form content became slightly more common. In November 2025, 20% of articles contained at least 2,500 words. By October 2026, that had increased to 26%.
The median article length also increased from 1,390 to 1,760 words.
Meanwhile, shorter articles became less common. The percentage containing fewer than 800 words fell from 17.8% to 12.5%.
But article length differed considerably depending on what companies were publishing.
Comparison articles were among the longest, with a median of 2,587 words. Opinion articles were substantially shorter at 1,389 words, while how-to articles had a median length of 1,998 words.
In a separate comparison of older and newer B2B articles, the average introduction before the first main heading decreased from 308 to 222 words.
So while longer articles became more common, newer writing also showed a tendency to get to the main subject sooner.
30% of articles had no image, video or table
Of the 10,172 articles, 30.2% contained no editorial image, video or table. Even among posts longer than 1,500 words, 65% had a stretch of at least 1,000 words without one.
In a review of images, 63% were screenshots, around 15% were diagrams, and only about 1.3% were data charts.
There were big company-level differences. Of 245 eligible publishers in the visual comparison, 56 illustrated at least 90% of their selected posts. Another 43 illustrated fewer than 20%.
Another 43.3% of articles had no counted editorial image. Among posts under 800 words, 49% were text-only. Video appeared in about 10% of articles, and roughly 60% of the embedded videos used YouTube.
Tables were especially common in comparison posts, appearing in approximately 73% of them. By contrast, 41% of opinion articles had no counted visual. Six of the 245 companies reviewed had no editorial illustrations in their selected posts.
Image descriptions also varied. Around 14% of images had generic alt text, 44% had descriptive alt text, 24% were missing it, and 10% had captions.
Why so many B2B articles look alike
27% of ‘What is X?’ articles followed the same four-part structure
In this research, 27% of eligible “What is X?” explainers followed a four-part structure:
- Introduction of a subject.
- Explains why it's important.
- Provides some guidance.
- Finishes with FAQs or a conclusion.
Looking at the individual sections, 56% of these explainers started with a definition, while 65% ended with FAQs or a conclusion.
The formula wasn't limited to articles explaining terminology. Among articles containing at least three main headings, 7% followed the same four-part structure, with articles from 146 of the 254 companies using it.
Individual elements were even more common. Some 31% contained a definition heading, while 29% included a heading about steps or how to do something.
The subject might be employee engagement, project planning or customer retention, but the article often followed a familiar sequence:
What is X? > Why does X matter? > How do you do X? > FAQs or conclusion.
Competitors' headings were seven times more similar
I also compared the heading structures of articles published by competing companies.
Among 2,522 pairs covering the same subject for the same search purpose, heading wording was roughly 7x more similar than in randomly selected articles from competing companies in the same sector.
The body text was only around 3x more similar.
This suggests that competitors writing about the same subject often resemble one another more in their headings than in the language used throughout the article.
The similarities weren't limited to articles covering the same topic. Companies also used similar heading structures for different subjects, often following the familiar sequence of definition, explanation, steps, and conclusion.
The content itself sometimes overlapped considerably.
The comparisons suggest that articles can use different wording while repeating many of the same recommendations.
How similar was competitors' writing?
A small sample found that 760 pairs of articles from different companies used unusually similar words and phrases. In a separate competitor comparison, 29% of pairs covering the same subject used more similar wording than 99% of randomly chosen pairs from their sector.
Put simply, many competing guides sounded alike, although similar words don't necessarily mean the same advice.
Product comparison articles stood out: 31% had a close wording match with an article from another publisher. That fell to 15% for listicles, 12% for guides, 9% for how-to posts, 9% for examples and templates, 5% for research articles, and 3% for opinion pieces.
These percentages show how similarly articles were worded, not how many gave identical information.
In a deliberately selected group of competing guides covering closely related subjects, 19 of 23 pairs repeated at least half of the shorter article's main points. For example, if one guide made ten recommendations, at least five also appeared in the competing guide.
How B2B writing is changing
AI writing has changed, including the em dash
One of the clearest changes in the writing was the decline in em dashes. They appeared in 66.7% of articles in November 2025, compared with 36.6% in October 2026.
The timing is key. In November 2025, ChatGPT became better at following instructions to avoid em dashes. This was confirmed by Sam Altman himself. So the drop in em dashes, combined with this timing, is not just a coincidence but a strong sign of content being written by AI.
I also compared newer articles with B2B writing published before ChatGPT. For example, the use of “We” fell from 15.4 to 6.9 mentions per 1,000 words. “You” fell from 26.1 to 18.1, and “I” from 4.2 to 1.7.
At the same time, abstract-noun patterns increased from approximately 3.8 to 5.0 per 100 words. The “rather than” construction rose from 2.4% to 8.9% of articles in the historical comparison.
I looked beyond the familiar AI writing clichés
I wanted to see whether B2B writing was changing in ways that resembled different generations of AI models.
To investigate this, I used a text-pattern analysis informed by Graphite's AI Tells research, which compared human-written content with output from models including GPT-4.1, GPT-6 Astra, Claude Opus 5.5 and Gemini 3.1 Pro.
The analysis examined several types of writing habits:
- Punctuation: Em dashes, colons, parentheses and other punctuation commonly associated with AI writing.
- Overused marketing language: Words and phrases such as "seamless," "robust," "streamline," "empower" and "unlock.".
- Formulaic sentence constructions: Phrases such as "not just X, but also Y," "rather than" and "this matters.".
- Transitions and introductions: Expressions such as "moreover," "furthermore," "let's dive in" and "in today's fast-paced world.".
- Writing style: Sentence length, vocabulary, abstract language, and the use of "we," "you" and "I.".
I also compared the newer articles with B2B writing published before ChatGPT to see which characteristics had changed and which were already common.
Stock marketing phrases fell by 63%
One of the clearest changes wasn't an increase in AI-generated language, but the disappearance of phrases that had become associated with it.
The use of 42 tracked marketing clichés fell from 1.31 to 0.49 occurrences per 1,000 words between the early and late publishing periods, a 63% decline.
Measured by how often they appeared per 1,000 words, terms including "seamless," "robust," "streamline," "empower" and "unlock" were used 65% to 76% less often in the later articles.
The formulaic invitation to "dive in" or "explore" also became less common, appearing in 7% of earlier articles compared with 3% of later ones.
Of 226 publishers tracked during the year, 77% reduced their use of em dashes. And 148 publishers reduced both their em-dash use and their stock marketing phrases.
There was a substantial change in how B2B companies were writing, including the removal of language that marketers themselves had used for years.
I also looked at how often articles used familiar marketing phrases and predictable sentence structures, such as "in today's fast-paced world," "let's dive in," and "it's not just X, it's Y."
36% of articles contained at least one of the nine types of repetitive language I tracked. Only 1.4% contained three or more different types, with examples appearing at 70 companies.
This language became considerably less common during the year. For every 100 articles, the analysis found 64 uses of different tracked writing-pattern types in the first six months, compared with 31 in the final six months.
Interestingly, companies that relied more heavily on these familiar phrases were also more likely to follow the standard explainer structure: definition, benefits, guidance, and conclusion.
Companies using more stock marketing phrases were also more likely to follow familiar article structures. The patterns are consistent with AI-assisted writing, although they don't prove which individual articles used AI.
Newer AI-associated phrases became more common
While older clichés declined, other patterns associated with newer models became more prominent.
Certain words and phrases that appeared frequently in tests of newer AI models also became more common in B2B writing. The share of articles showing a high concentration of these patterns increased from 6.1% in pre-ChatGPT writing to 14.2% in newer content.
Meanwhile, older GPT-4-era vocabulary patterns declined from 8.5% to 5.1%.
One particularly noticeable change was the increased use of statements emphasizing importance. The specific "this matters" phrase family appeared in approximately 2.4% of older articles, compared with 25.7% of newer ones.
There was also a move away from traditional generic introductions. The familiar "in today's fast-paced world" opening appeared in 2.1% of pre-ChatGPT articles but only 0.7% of the newer content.
55% of newer articles fell outside the earlier B2B writing-style range
When I compared older and newer articles from the same publishers, 55% of the newer articles scored higher on the combined writing-style measure than 95% of their pre-ChatGPT articles.
Sentence structure changed too. The average sentence became shorter, falling from 17.8 to 15.3 words. Average word length increased from 5.05 to 5.39 characters, while vocabulary became more varied.
Altogether, 14 of the 21 writing patterns examined during the study moved in the same direction as newer AI models.
Another important finding here:
Seven of the twelve comparable writing characteristics were already closer to model-style writing than to the general human-writing reference before ChatGPT existed.
Some of the language people now associate with AI, including abstract nouns, generic subjects, and formulaic explanations, was already common in B2B marketing.
Not every characteristic moved toward model output. Sentence-length variation, for example, remained closer to the human-writing reference.
What this means
The research shows a substantial change in B2B SaaS writing. Older marketing clichés are disappearing, newer language patterns are appearing, and companies are using less personal language than they did before ChatGPT.
To be clear, this part used common patterns associated with AI against the article dataset - it doesn’t definitively mean “AI wrote this”.
For B2B marketers, the more important question is whether those changes are making the writing better.
Removing "seamless" or avoiding em dashes won't give an article a distinctive brand voice. Neither will replacing one familiar phrase with another.
What makes B2B writing worth reading is the information and perspective behind it: specific observations, original examples, genuine expertise and explanations that go further than the advice every competitor already publishes.
What B2B articles use as evidence
56% of articles contained numbers. Original evidence was much rarer.
56% of the articles included percentages, ratios, or other numerical claims.
But these weren't necessarily original statistics or research findings. They also included examples, targets, and figures repeated from other sources.
Only 18% of numerical statements had a clearly attributed external source. One-third of posts with at least three detected numerical claims had no detected attribution for any of them.
Original research was harder to find. About 9% of articles showed signs of first-party data, 3% described a method or sample, and approximately 1% had both.
I checked 45 numerical claims against the evidence behind them. Just 13 were supported, and only seven could be confirmed directly from the sources I opened. The other six relied on search snippets or second-hand evidence.
Ten claims were partly supported or exaggerated, one wasn't supported by its linked source, and 15 couldn't be verified. The remaining six were examples or product specifications rather than research findings.
In one article, a company presented an ROI estimate covering several years as though the entire return happened in the first year. The source was real, but the way the number was used gave readers a misleading impression.
Articles packed with statistics were also less likely to identify where those numbers came from. In posts containing just one or two numbers, 42% of those numbers had an identifiable source. That fell to 23% in posts containing 21 or more.
Of the 3,531 articles containing at least three numbers, 1,176 had no detected source for any of them.
Interestingly, the most frequently repeated statistics were more likely to have links. Around 79% of appearances of the 30 most-repeated numbers had an external link, compared with roughly 25% of numerical claims overall.
But a link didn't guarantee accuracy. I found an "up to 60% faster hiring" claim repeated by two publishers without a clear source, alongside Pinterest audience figures that referred to different reporting periods.
The research being cited was also concentrated among a handful of familiar names. Of 2,384 links to research and advisory firms, 71% referenced just five organisations: Gartner, McKinsey, Gallup, SHRM and Deloitte.
In the separate comparison of pre-ChatGPT and newer articles, the share showing signs of first-hand research fell from 26% to 18%. Meanwhile, articles mentioning named customers increased from 8.5% to 15%.
Authors and content updates
Of the 10,172 articles, only 42% visibly named an individual author. Another 10% had a byline, such as a company team name, while 32% identified an author only in the page's hidden metadata.
The remaining 16% had neither. Just 2% named someone responsible for editing or reviewing the content.
I also looked at how up-to-date the articles appeared to be.
Nearly 10% mentioned 2022 or an earlier year in their stats or content. That doesn't necessarily mean the information was outdated, but it's worth checking when an article relies on older statistics or research.
Meanwhile, 4,753 articles had been modified since their original publication date and another 2,238 didn't report a modification date at all.
I also found 220 articles with a year in the headline that was newer than their recorded publication year. For example, an article published in 2024 - now titled ‘Best CRM Software for 2026.’ A common tactic to retain link equity for listicles that get refreshed every year to stay relevant.
How B2B articles lead readers towards products
Internal links are everywhere. Customer evidence isn't.
The median article had seven internal links and one external link. Some 37% had no counted external links at all.
Only 2% of classified editorial internal links pointed to customer stories. Among 5,567 sections about evaluation or implementation, 17% linked to documentation or product pages.
So companies are linking to other content, but they aren't always connecting an explanation to the material someone might need next.
About 9% of articles had no counted internal link. Of 50 inspected related-reading links, 82% fit the topic. Only about 0.8% of counted anchor text used generic wording such as “click here.”
Some 43% of posts also linked from the main body to a product, pricing, signup, or similar commercial destination. In a small link check, 11 of 12 such links led to product or feature pages, rather than necessarily asking for a demo.
There was a more unusual pattern in 41 posts from one company dated April 2026: citations routed through Google AI-search redirect URLs instead of linking directly to the source. One checked redirect ended at a homepage that didn't contain the stated number.
This suggests that AI was used for the content and citations/data copied directly without being fact-checked.
Product promotion often appears before the final CTA.
About 49% of articles were detected as having an editorial CTA in the body.
Of the CTAs found, 47% appeared in the final tenth of the article. The first appeared around 79% of the way through.
Among the offers identified, about 26% promoted demos, 20% trials, and 15% downloads.
But companies don't wait until the end to promote themselves. Around 43% of articles had a link in the main text to a commercial destination, such as a product or pricing page.
Some brands repeat the same offer in almost everything they publish. 34 companies had a detected CTA in at least 95% of their selected articles.
The vendor comparison findings showed that in 23 selected software comparisons that included the publisher's product, 19 put it first. Twenty-two acknowledged a competitor advantage, but only five mentioned a weakness of their own product.
Posts without a genuine in-body CTA fell between the ranges of 40% to 53%. Around 6% of detected CTAs appeared before the first H2.
Forty of 204 eligible publishers reused exactly the same CTA on at least 80% of posts containing a CTA, and publisher habits accounted for approximately 39% of the variation in whether one was present.
Vendor roundup headlines increased from 3.3% to 11% in the older-vs-newer company comparison. In the sample, 19 of 23 eligible vendor comparisons put the publisher first (83%), 22 credited a competitor's advantage (96%), and five admitted an own-product weakness (22%).
CTA use declined slightly during the year. The differences between publishers were much bigger than the differences between educational and product-focused posts.
Where B2B SaaS blogging goes from here
The increase in FAQs and tables is real. So are the changes in writing style, the lack of first-party evidence in many articles, and the heavy use of familiar sales offers.
I don't think the solution is to stop using tables, FAQs or product links. They can all improve an article when there's a good reason for them.
But I'd stop treating formatting as evidence that the content is good. Check the statistics. Show how something is done. Explain the limitations. Use examples a competitor couldn't copy by changing the brand name.
With AI making content production faster, those details deserve more attention than another template.
Study note: The analysis covers 10,172 selected articles from 254 B2B SaaS companies in five sectors, with no more than 43 per company, dated November 2025 to October 2026. Some pages were modified after their publication dates. The older-versus-newer writing comparisons covered 76 companies, with a stricter check on 33; 105 distinct articles were read for detailed comparisons. These smaller studies are kept separate from the 10,172-article percentages.