B2B SaaS Content in AI Search: What 10,382 Keywords and 350,000 Articles Show

Written By Ryan Edwards, CAMINO5 | Co-Founder

New B2B SaaS research shows statistics, expert quotes, and cited sources drive higher Google rankings and more AI citations.

Organic search accounts for about 83% of traffic for B2B SaaS. Every other channel contributes single digits. For B2B, especially in discovery, that makes AI search the largest potential disruption, not social and not UGC like Reddit.

Most studies are general. They talk about every category, all of AI search, all of organic.

Backlinko looked at 11.8 million Google results from every category, recipes included. Princeton used 10,000 queries, mostly academic. SE Ranking studied 2.3 million pages but never said which niches. Ahrefs looked at 55.8 million AI Overviews worldwide.

Every vertical behaves differently. Here we take a deep dive into just B2B SaaS.

Most B2B SaaS content is not built for how buyers now find answers and the gap is not technical. It is editorial. Statistics, named experts and cited sources are what both Google and AI engines reward, and four out of five B2B SaaS resources simply don't have them.

To be clear, you do need to have your technical elements in place: your schema, your topic, and your topical authority. At the end of the day, it is the stats, the named experts, and the cited sources that make you the expert and get you discovered.

 

Method

We collected roughly 138,000 raw B2B SaaS keywords from four sources: G2 category pages across 52 categories, the blogs of leading SaaS companies, related-term expansion through a commercial keyword provider, and query templates ("X vs Y," "best X software," "how to Y") applied across all categories. Every keyword required 50+ monthly searches. After filtering, 10,382 keywords remained.

For each keyword we collected Google's top 20 organic results and queried four AI engines: ChatGPT with web search, Claude with web search, Perplexity and Google AI Overviews. We then extracted 17 content features from every article using AI extraction and automated parsing, validated against 200 manually reviewed articles at 85.1% to 100% per-field accuracy.

Methodology

What we looked at, and how much of it.

Table 1 Study corpus
Corpus Count
Keywords 10,382 Across 52 categories
Google result positions 194,821
Unique Google articles, top 20 143,117
AI article references collected 487,982
Unique AI-cited articles 301,604
Articles with full feature extraction ~350,000 Across both the Google and AI-cited sets
Table 2 Keyword distribution by intent
Intent category Keywords Example
Question-format ~2,480 “how to automate onboarding”
Comparison ~2,350 “HubSpot vs Salesforce”
Best-of ~2,160 “best project management software”
Use-case ~1,290 “scheduling app for restaurants”
Feature ~1,230 “CRM email tracking”
Category ~880 “content marketing platform”
Total 10,382 Six intent categories

Bars are scaled to the largest category, not to the total. Category counts are rounded and sum to within eight keywords of the corpus total.

 

Summary of findings

  • Most B2B SaaS articles have no stats, quotes or cited sources. The few that do rank higher on Google and get cited more often by AI search across the board.

  • Word count has no ranking effect inside Google's top 10 for SEO. This matters for one simple reason: it gives you more flexibility in making content that is both AI-search and SEO-friendly.

  • Earned media gets 61% of AI citations. Your own content gets 29%.

    When you average all categories and verticals, brand content only gets 8%, but in B2B SaaS your own site gets cited three times more. Tactics like blogs, product pages, how-tos, and explanations still count.

  • Keyword-in-URL and keyword repetition show no positive effect on ranking or AI citation.

  • Schema makeup does not have an impact on ranking for SEO, but it does drive AI citations, although not as much as it used to. What is interesting is that schema does have a measurable impact on click-through rates For both AI search and SEO

  • AI Overviews trigger on most informational B2B SaaS queries. Comparison keywords hit 87%.

 

Finding 1: credibility signals separate winners from losers

Content features

Quotes and statistics separate cited from ignored. The checklist does not.

Table 3 Content feature baselines across article groups
Feature All articles Google position AI citation Lift
1–5 11–20 Cited Not cited
Strong separation
Mean expert quotes 0.6 1.5 0.4 1.6 0.2 8.0x
% with 1+ expert quote 21% 44% 25% 52% 12% 4.3x
Mean statistics 1.9 3.6 1.9 4.2 1.2 3.5x
Mean source citations 3.2 5.3 3.2 6.2 2.3 2.7x
% with 3+ statistics 29% 63% 36% 64% 28% 2.3x
Moderate separation
Mean section count 9 11 8 12 8 1.5x
FAQ section 19% 22% 16% 24% 17% 1.4x
Mean word count 1,392 1,642 1,284 1,690 1,305 1.3x
Readability grade 10.5 9.9 10.7 9.6 10.8 −1.2
No separation
Keyword in title 73% 74% 71% 69% 74% 0.9x
Schema markup 71% 72% 70% 69% 72% 1.0x

Lift is the AI-cited value divided by the not-cited value, computed from the columns shown. Readability is reported as a grade level, so a lower figure means simpler prose and the difference is shown in grades rather than as a ratio. Rows are ordered by separation strength, not as they appeared in the source table. These are group averages, not controlled effects.

  • Only 21% of B2B SaaS articles include a named expert quote. Only 29% include three or more statistics. In Google's rankings 44% (expert quote) and 63% (3+ stats).
    Among AI-cited articles, 52%  (expert quote) and 64% (3+ stats).

  • Princeton's controlled experiment supports the direction. Quotations improved AI visibility by 28% to 43%, statistics by 23% to 33%, source citations by 13% to 28%. The new information is the baseline. Four out of five B2B SaaS articles have no expert quote.

  • Reading level works differently here. SE Ranking says grade 6 to 8 wins when you average all categories and verticals. B2B SaaS runs higher. The average article reads at grade 10.5. Google's top five read at 9.9. The articles AI cites read at 9.6. Grade 9 to 10 is your target.

  • Keyword-in-title and schema markup are flat across every group, at 69% to 74% and 69% to 72%. Neither moves position or citation. Both are table stakes.

 

Finding 2: the Google position gradient

Position gradients

Evidence thins as you move down the page. Density stays put.

Table 4 Content features by Google position
Feature Google position band Trend 1–5 vs 16–20
1–5 6–10 11–15 16–20
Declines steadily with position
Mean expert quotes 1.5 0.9 0.6 0.2 7.5x
Mean statistics 3.6 2.8 2.2 1.6 2.3x
% with 1+ expert quote 44% 34% 28% 22% 2.0x
Mean word count 1,642 1,463 1,338 1,230 1.3x
Readability grade 9.9 10.3 10.6 10.8 −0.9
Declines, but runs against the usual reading
Mean article age, months 23 22 18 14 1.6x
No gradient
Keyword density 1.3% 1.2% 1.4% 1.3% 1.0x

Sparklines are scaled to each row's own range, so they show the shape of the change and not its size. Readability is a grade level, so its rising line means denser prose further down the page and the difference is shown in grades. Position bands are averages across the corpus, not the same articles tracked over time.

  • Word count shows a gradient across the top 20 and flattens inside the top 10. Top-five articles average 1,642 words against 1,230 at positions 16 to 20, but the gap to positions 6 to 10 is 179 words. Backlinko found zero within-page-one correlation across 11.8 million results. Length is a symptom of thorough coverage, not a cause of ranking.

  • Expert quotes show the steepest gradient, from 44% at positions 1 to 5 to 22% at 16 to 20.

  • Age confirms incumbency. The mean top-five article is 23 months old against 14 months at positions 16 to 20. Only 9% of top-20 articles were under 12 months old, against 13.7% on the general web. This category rewards maintenance over publishing volume.

  • Keyword density is flat at 1.2% to 1.4%. Repetition is not a strategy.

 

Finding 3: AI Overview competition by query type

AI Overviews

Comparison queries almost always trigger one. Then the clicks halve.

Table 5 AI Overview trigger rates by query type
Query type Trigger rate
Comparison 87%
Question-format 83%
Category 73%
Best-of 72%
Use-case 58%
Pricing and evaluation 42%
Feature-specific 41%
Transactional 8%

Bars run on a fixed 0 to 100% scale with a mark at the halfway point, so lengths are comparable across rows and against the full scale.

Table 6 Click-through rate by AI Overview scenario
Scenario Click-through rate Against no overview
No AI Overview 3.35% 33,500 clicks per 1M impressions Baseline
AI Overview, brand cited 2.07% 20,700 clicks per 1M impressions −38%
AI Overview, brand not cited 0.94% 9,400 clicks per 1M impressions −72%

Clicks per million are the same figure as the click-through rate, restated at scale. Being cited recovers roughly 2.2 times the clicks of not being cited, but still lands 38% below a result page with no overview on it.

  • Comparison and question-format queries face near-universal AI Overview competition.

  • Only transactional queries are unaffected. Citation recovers about 62% of pre-AI-Overview click volume. No citation recovers 28%.

  • Citation is loss reduction, not growth. Agarwal and Sen's randomized field experiment puts the aggregate effect at a 38% drop in organic clicks and zero-click searches rising from 54% to 72%.

 

Finding 4: what AI engines reward

Same keyword, different outcome

Cited articles carry evidence. And they live somewhere else.

Table 7 AI-cited versus non-cited articles for the same keywords
Feature AI-cited Not AI-cited Difference
What the article contains
Mean expert quotes 1.6 0.2 +1.4
% with 1+ quote 52% 12% +40pts
Mean statistics 4.2 1.2 +3.0
Mean word count 1,690 1,305 +385words
Readability grade 9.6 10.8 −1.2grades
Where the article lives
On a review platform 19% 9% +10pts
Earned media 61% 45% +16pts

Both groups answer the same keywords, so the comparison holds the query constant. Readability is a grade level, so the negative difference means the cited articles read more simply.

Table 8 Domain composition by context
Domain type Google top 20 AI-cited General web
Earned media
Brand-owned
Reference

The first two columns are shares of the same set and each totals 100%. The general web column is drawn from a separate measurement, does not include a reference figure, and so does not total 100%; treat it as background rather than as a third comparable column.

  • The gap between cited and non-cited articles for the same keyword is 40 percentage points on expert quotes and 3.0 statistics per article, the two largest content-feature differences in the study.

  • Domain mix is where B2B SaaS separates from the published research. Brand-owned content holds 29% of AI citations here against roughly 8% on the general web. Earned media holds 61% against 82% to 85%.

  • Review platform domains account for 19% of AI-cited articles against 9% of non-cited. LinkedIn is now the fifth most-cited domain on ChatGPT and the most-cited for professional queries across all six major AI platforms.

  • Citation is unstable. Profound found 40% to 60% of cited sources change month to month, highest on AI Overviews at 59.3% and lowest on Perplexity at 40.5%. One ChatGPT entity update in October 2025 removed 31% of tracked brand visibility across more than 85% of tracked brands.

 

Finding 5: what has no effect

Null results

Five things the data would not support. Including two we recommend.

Every table before this one reports something that moved. This one reports what did not, including the levers that appear on most technical SEO checklists.

Table 9 Features with no measurable positive effect
Feature Google ranking impact AI citation impact
Schema markup No effect No gradient across positions. No effect Present at similar rates whether cited or not.
URL length Negligible Too small to act on. No effect No separation between groups.
Keyword in URL No effect None within the top 20. Inverse Cited pages carry it less often, though the relationship is likely confounded.
Keyword repetition No effect None within the top 20. Negative, about −9% Repetition reduces citation probability. Princeton
FAQ schema Not measured Outside the scope of this study. Weaker than the content An actual FAQ section separates cited from uncited articles; the markup around it does not.

No effect means no measurable separation in this corpus, which is not the same as proof that a feature does nothing. Rows marked in gold report a measured negative or inverse relationship rather than an absence of one.

  • Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026 and found no lift in AI citations. AI engines read visible page content, not metadata.

  • Schema still helps Google interpret a page for rich results and entity disambiguation, but it does not make an engine cite you.

 

Finding 6: Google and AI draw from separate pools

Citation overlap

The engines barely agree with each other. Or with Google.

Ranking on Google buys a minority stake in AI citations, and being cited by one engine says almost nothing about the others.

Table 10 Google-to-AI and cross-engine citation overlap
Comparison Overlap
Google against AI, measured both ways
Of Google top-20 URLs Also cited by AI 30%
Of AI-cited URLs Also in Google’s top 20 14%
Engine against engine
Perplexity and AI Overviews 17%
ChatGPT and Perplexity 10%
ChatGPT and Claude 8%

Bars run on a fixed 0 to 100% scale with a mark at the halfway point. The first two rows describe one relationship measured from each side: the difference between them means the AI-cited pool is the larger of the two sets. Overlap figures depend on how many URLs each engine surfaces per query, so add the method and the measurement window to the source line.

  • Only 14% of AI-cited URLs appeared in Google's top 20. Reversed, 30% of Google's top-20 articles were cited by at least one AI engine. The asymmetry is a size effect: the AI pool holds 301,604 unique articles, the Google pool 143,117.

  • Ranking on Google gives you roughly a one-in-three chance of AI citation, and six of every seven AI-cited articles do not rank on Google at all. Cross-engine overlap runs 8% to 17%. There is no single AI channel to optimize for.

 

Where B2B SaaS differs from the general web

Vertical divergence

B2B SaaS does not behave like the open web. Owned content still counts here.

Prior research is drawn from general-web corpora. Where our findings diverge, the vertical is the reason.

Table 11 B2B SaaS findings against general-web prior research
Finding General web B2B SaaS The read
Diverges sharply
Reddit share of AI citations ~40% AbsentNot a factor Engines prefer structured comparison over forum discussion for these queries.
Brand content share of AI citations ~8% 3.6x higher29% Owned content is roughly three times more viable here than on the open web.
Diverges moderately
Earned media share of AI citations 82–85% Lower61% Vendor-specific queries pull citations back toward the vendor.
Review platform citation lift 3–3.5x Lower2.1x A smaller multiple, but still material.
Top-20 pages under 12 months old 13.7% Lower9% Stronger incumbency. New pages displace old ones less often.
Optimal readability Grade 6–8 2–3 grades higherGrade 9–10 A technical audience tolerates more complexity.
Holds
AI and Google source overlap 11.9% In line14% Consistent with prior research.

The read column is interpretation, not measurement. General-web figures come from prior published research rather than from this corpus, so the two columns are compared across studies and not within one dataset.

 

What to do with this

What to do about it

Seven actions, ordered by the evidence behind them.

One of these has causal support. The rest rest on association, which is worth acting on but worth naming.

Table 12 Priority actions and supporting evidence
Action Evidence
Add statistics, named expert quotes and cited sources to every article Causal The only action here with causal evidence behind it. Cited articles average 4.2 statistics and 52% carry at least one quote, against 1.2 and 12% for uncited.
Fund earned media alongside owned content Association 61% of AI citations go to earned media against 29% to owned.
Build and maintain G2, Capterra and TrustRadius presence Association 19% of cited articles sit on a review platform against 9% of uncited.
Run Google and AI as separately measured channels Association Source overlap runs 14% against Google and 8–17% between engines. One channel does not stand in for the other.
Structure articles in clear, self-contained sections Association Cited articles average 12 sections against 8 for uncited.
Write for a technical reader, not a general one Association Cited articles read at grade 9.6 against 10.8 for uncited, and this vertical sits two to three grades above the general-web optimum.
Update existing articles before publishing new ones Weakest inference The mean top-five article is 23 months old and only 9% of top-20 pages are under a year. That shows incumbency, not that updating causes it.

Every figure here traces to a table above. Association means the feature separates cited from uncited articles in this corpus; it does not establish that adding the feature produces the citation.

 

Bottom line

Most B2B SaaS content is not built for how buyers now find answers. The gap is not technical. It is editorial. Statistics, named experts and cited sources are what both Google and AI engines reward, and four out of five articles in this category still do not have them.

 

Sources

Ryan Edwards, CAMINO5 | Co-Founder

Ryan Edwards is the Co-Founder and Head of Strategy at CAMINO5, a consultancy focused on digital strategy and consumer journey design. With over 25 years of experience across brand, tech, and marketing innovation, he’s led initiatives for Fortune 500s including Oracle, NBCUniversal, Sony, Disney, and Kaiser Permanente.

Ryan’s work spans brand repositioning, AI-integrated workflows, and full-funnel strategy. He helps companies cut through complexity, regain clarity, and build for what’s next.

Connect on LinkedIn: ryanedwards2

Next

What Actually Earns Citations in AI Search