Where AI Answers Get Their Sources for B2B Software, and Why Only 10% Point to Your Site

In a 2026 study of 57.2 million AI citations, only 10.15 percent pointed to brand-owned websites. Reddit made up 20.8 percent of external citations, and it was higher for unbranded queries. For B2B software teams, that means a content plan limited to your own blog covers about a tenth of what AI actually cites.

This article explains what the data says, where it is weak and how to adjust a content plan. It is written for SaaS marketers and founders who run their own content program. Cekpr covers search and AI visibility for B2B software, and this is one of the clearest signals that publishing more on your own site is not the same as showing up in AI answers.

What Did the 57.2 Million Citation Study Find?

Foundation Marketing and AirOps published a report on brand visibility in AI citations, last updated May 5, 2026. It tracked 50 brands across seven B2B verticals for 60 days, from December 2025 through February 2026. The platforms were ChatGPT, Gemini, Perplexity, Google AI Mode and Google AI Overviews.

The dataset covered 5.1 million responses and 57.2 million citations. Each brand got 100 prompts, 65 percent unbranded discovery questions and 35 percent branded validation questions. That split matters, because buyers form shortlists with unbranded queries.

The headline number is 10.15 percent. That is the share of citations that linked to brand-owned domains. Brand content was missing from more than two-thirds of AI responses, even when the brand was discussed.

Why Does the Prompt Type Change the Result?

When a user asks about a brand by name, the brand's own site shows up often. In the study, brand-owned content appeared in 77.6 percent of branded-prompt responses. That is the validation stage, where a buyer checks facts on a vendor they already know.

Discovery looks different. For category-level prompts such as "best project management software for agencies," only 2.2 percent of citations were brand-owned. The model leans on other sources to compare vendors, because it has little reason to trust each vendor's own claims.

That gap is the strategic point. Branded visibility is easier and matters less for new buyers. Category visibility is where shortlists form, and your own blog covers little of it.

Which Sources Does AI Lean On Most?

Reddit led. It accounted for 20.8 percent of external citations overall, rising to 30.9 percent on unbranded queries, and it was the largest source in six of seven verticals. YouTube followed at 13 percent and LinkedIn at 11 percent.

Help and support documentation made up about 8 percent. G2 contributed about 4 percent. The report estimates that about 39 percent of the citation footprint is manageable, meaning on-domain content, help docs, community, review sites, video and earned media. The other 61 percent is unmanaged third-party space.

The report also says there is no universal playbook. Fintech leaned on NerdWallet and Bankrate, DevOps on GitHub, Medium and dev.to, and productivity tools on YouTube. Check your own category before you copy anyone's plan.

How Reliable Are These Numbers?

Treat all citation studies with some caution. Vendors that sell AI visibility tools run most of them, query sets differ, and platforms change quickly. The Foundation and AirOps report is large and specific, but it covers a window that ended in February 2026.

A second source, a GrackerAI study of 10,000 citations, reported Reddit at 40.1 percent and vendor sites below 4 percent. Its methodology has gaps, though, including percentages that do not reconcile and thin sampling detail. Use it as a loose indicator that the direction matches, not as a precise rate.

Both studies agree on the direction: owned sites are a minority of what AI cites. Neither proves how a given brand will fare, so test your own prompts. Our guide to measuring AI search visibility in Search Console covers the on-site side of that tracking.

What Does This Mean for a Content Plan?

Keep your owned content strong, and extend the plan beyond it. A balanced plan has four parts: owned pages, documentation, community presence and earned mentions. Each serves a different stage of the buyer's search.

Owned pages should answer the questions buyers ask, with clear claims and sources. Google's guidance on helpful, people-first content is a good baseline for that work. It asks whether a page offers original information and whether a reader would leave satisfied.

Documentation deserves more attention than it gets. At 8 percent of citations, help docs are cited more than G2 in this data, and you control them fully. Keep them current, specific and free of marketing language.

Community and review spaces need honest participation. Answer questions in the places buyers ask them, disclose your affiliation and avoid stuffing links. Communities punish promotion, and AI systems cite useful threads, not ads.

How Do You Build a Plan Step by Step?

You do not need a large team to start. Use these steps to turn the data into a plan:

Accuracy comes first. If an AI answer gets your pricing or features wrong, trace it to the page it used and fix that page. Our guide on fact-checking AI-assisted content offers a review process you can apply to your own documentation.

Technical shortcuts will not replace this work. Our look at whether schema markup increases AI citations shows that a single markup change rarely moves visibility. Content and sources do.

How Should Cekpr Readers Read Vertical Differences?

The study's most useful caution is that sources vary by vertical. A developer tool and a finance product do not share a citation mix. That means a team in HR software cannot assume that what works for DevOps will help them.

Start with your own category. Pull the questions your buyers ask and see which domains repeat. If a handful of review sites, a few YouTube channels or two community forums dominate, those are your priority targets.

Cekpr's editors suggest a short list of three to five recurring sources per category. Study what makes those sources useful to the model: clear comparisons, specific numbers, recent dates and plain language. Then find honest ways to appear there, such as customer reviews, expert answers and product walkthroughs.

Why Do Product Facts Matter So Much?

AI answers repeat facts about your product, and they often get them wrong. Pricing, integrations, plan limits and security claims are the usual errors. When the model learns them from a stale forum post or a competitor's comparison, the mistake spreads.

Make the facts easy to find and hard to misread. Put pricing, integrations and limits on dedicated pages with dates. Keep one source of truth, and update the help center when features change. Add a short "last updated" line so models and readers can tell which version is current.

Then monitor. Ask the main AI platforms the same product questions each month, and log where the answers are wrong. Treat each error as a content bug: find the source, fix or answer it, and recheck in a few weeks.

This work is slow and unglamorous, but it compounds. A clean set of facts across your site, docs and profiles is the foundation for every other tactic.

What Mistakes Should You Avoid?

Do not flood forums with promotional posts. Moderators remove them, and users notice. Do not chase one study's numbers as if they were fixed rates, since platforms change. Do not ignore the help center because it feels unglamorous.

Do not measure only clicks. AI answers often satisfy a question without a visit, so brand mentions and branded search growth matter too. Track both.

A final point on timing: citation patterns shift as platforms update, so schedule a quarterly review of the plan, not a one-time project. What worked in the winter data may look different by next spring.

Conclusion

Brand-owned pages are only about a tenth of what AI cites for B2B software. Strong owned content still matters, but discovery happens elsewhere. Audit the sources AI uses in your category, fix what you control and earn a place in the communities and reviews buyers trust.