SaaS AI Search Optimization: New Data Shows 84% of Citations Come From Outside Your Website

New research on saas ai search optimization from SEO consultant Aleyda Solis, published July 16, 2026, shows that 84% to 93% of AI citation weight for SaaS brands comes from third-party sites, not from your own domain (Aleyda Solis, July 2026). The study analyzed 15 leading SaaS brands across three US subverticals using Semrush citation data for ChatGPT and Google AI Mode plus Similarweb AI referral data. The takeaway is not that your website matters less. It is that your website is only one leg of a three-legged stool, and most operators are only funding one leg.

If you run marketing at a B2B SaaS company or an agency that serves one, the data below tells you where to spend the next quarter. It also explains why the pages you keep polishing may not be the ones AI is citing.

What the New SaaS AI Search Optimization Data Shows

Aleyda studied 15 brands: Salesforce, Zoho CRM, Calendly, HubSpot, and Apollo in CRM and Sales; Atlassian, Notion, Slack, Dropbox, and Figma in Project and Work Collaboration; and Intuit, Zoho Finance, Xero, Wave, and Ramp in Accounting and Finance. She measured three layers: brand-owned pages earning citations, external sources shaping the answers, and pages receiving measurable AI traffic afterward.

The headline number: across every subvertical and both AI platforms, third-party sources generate 84% to 93% of citation weight. On ChatGPT, CRM and Sales sees only 7.9% owned citations. Project and Collaboration sees 8.1% owned. Accounting and Finance is the only category with double-digit owned share at 16.4%, and it still leaves 83.6% of citation weight on external sources. Google AI Mode is nearly identical, with owned shares of 6.6%, 7.8%, and 15.9% across the same three categories.

A separate July 2026 study from Writesonic found that 96% of AI citations across a broader sample point to third-party sources, up from around 80% just a few months earlier (Search Engine Journal, July 2026). Two independent measurements, same direction. This is not a rounding error. It is the new shape of AI visibility.

ChatGPT and AI Mode Are Not the Same Game

The second finding is the one that changes budget conversations. ChatGPT and Google AI Mode both cite external sources, but they cite different kinds of external sources, and they weight different parts of your own website differently. Running a single AI visibility program for both platforms will systematically underfund one of them.

On the external side, ChatGPT leans on structured written sources. Peer software vendors generate 25.4% of its external evidence share. Communities and forums, mostly Reddit and Quora, deliver 14.4%. Technology publications like TechRadar and PCMag sit at 7.7%. Review platforms like G2 and Capterra are at 6.4%. Video from YouTube is only 1.0% of ChatGPT’s mix.

AI Mode inverts the pattern. Video jumps to 23.0% of external evidence. Social platforms like LinkedIn, Facebook, and X move from 2.0% to 9.0%. Creator platforms like Medium and Substack rise from 0.4% to 3.0%. Technology publications collapse to 0.5%. If you have a content team that only publishes long-form written guides and pitches trade press, you are optimizing for ChatGPT and losing AI Mode. Our team walked through the ranking signal split in more detail in the top 6 AI search ranking signals for 2026, and the SaaS data now confirms what that piece predicted about platform-specific evidence.

Your Homepage Is a ChatGPT Asset, Not an AI Mode Asset

The owned-content patterns are even sharper. On ChatGPT, root URLs carry 14.7% to 17.7% of your citation weight. Corporate homepages sit at 13.7% to 16.9%. Help and documentation pages carry 19.7% to 26.6%. In other words, ChatGPT rewards canonical brand surfaces and reusable reference pages.

AI Mode does the opposite. Root URLs collect just 0.8% to 4.1% of AI Mode citation weight. Homepages sit at 0.4% to 4.0%. Instead, AI Mode concentrates 69% to 76% of its citation weight on URLs two to three path segments deep, where guides, templates, comparisons, and community content live. Editorial and guides carry 25.0% to 34.9% of AI Mode’s owned share, versus 15.5% to 24.9% on ChatGPT. Comparison pages more than double their weight in AI Mode.

The practical implication: if you have been sending PR and homepage optimization budget to lift AI visibility, you are helping ChatGPT and largely ignoring AI Mode. And if you have only been building deep guides and comparisons, you are helping AI Mode and largely ignoring ChatGPT. You need both layers, and they need to be internally linked so evidence pages lead to a commercial next step. This is the same pattern our team documented in the four-pass GEO content audit that diagnoses why AI Overviews quote Reddit instead of your page.

Grok Adds a Third Pattern: Live X Plus the Open Web

ChatGPT and AI Mode are not the only engines your buyers ask. Grok, xAI’s assistant, now sits inside X and answers buying questions there. It plays the citation game differently from both, so a program tuned only for the first two will miss it.

Grok runs two retrievers at once. One reads the open web the way every other engine does. The other, X Search, reads the live post graph on X directly, according to xAI’s documentation. No other major assistant has that native access. Evergreen questions get answered mostly from the durable web. Live questions get answered from what is being said on X right now.

That live feed is why every vendor selling a Grok tracker tells you to flood X with posts. The independent data says the opposite. According to Ahrefs, across 1.9 million US queries in June 2026, X was only Grok’s twelfth most-cited domain, a rounding error behind Reddit, YouTube, and Facebook. Grok has X built in, and X is still near the bottom of what it actually cites.

Grok’s most-cited domains (June 2026)	Mention share
Reddit	16.3%
YouTube	15.1%
Facebook	13.9%
x.com	1.4%

Source: Ahrefs, mention share across 1.9 million US queries, June 2026.

Grok also tends to say your name instead of quietly borrowing your page. Its unnamed-source rate sits at 22%, against 37% for ChatGPT and 52% for Perplexity, according to Search Engine Land. For a lead buyer, being named beats being linked, which makes Grok worth a dedicated play. We broke down exactly how to get cited by Grok, pool by pool, in a separate playbook.

The Pages Getting Cited Are Not the Pages Getting Traffic

The fourth pattern is the one that quietly breaks most reporting stacks. Cited pages and destination pages are different sets of URLs, and the gap widens as products become more operational.

In CRM and Sales, primary cited pages are product, docs, support, academy, and comparison URLs. Primary AI traffic destinations are product, pricing, homepage, app access, and account pages. About 42.7% of AI traffic in the top page sample lands outside cited URLs. In Project and Collaboration, cited pages are docs, templates, integrations, and workflow guides, but traffic lands on apps, workspaces, logins, and OAuth or MCP authorization flows. Fully 67.3% of AI traffic sits outside cited URLs. In Accounting and Finance, guides and calculators earn the citations, and pricing and calculators receive the clicks, with a much tighter 22.4% gap.

Read that carefully. AI cited URLs are not simply a subset of AI traffic URLs. They are a separate inventory that shows where AI systems reference information, even when those pages do not receive the measurable click. If your dashboard reports citations and referrals under one KPI, you are washing out the signal. Any serious saas ai search optimization program has to report them as separate, connected metrics with defined destination quality by category. The two-pass retrieval fix for Perplexity citations uses the same separation logic to reduce the citation-to-traffic gap for lead-gen brands.

What SaaS AI Search Optimization Looks Like By Category

The final finding is that no universal SaaS AI search optimization playbook exists. Each category has its own dominant user job, and AI systems reward the content that helps users complete that job. Copying another SaaS company’s editorial strategy optimizes for someone else’s answer space.

For CRM and Sales, the dominant job is “understand, compare, and adopt.” Owned assets to prioritize: product, support, academy, implementation, and comparison pages. External priorities: reviews, directories, peer vendors, publishers, and communities. Primary KPIs: demo, signup, pipeline, app access, and support resolution.

For Project and Work Collaboration, the job is “complete workflows and integrate.” Owned priorities: documentation, templates, integrations, utilities, and workflow guides. External priorities: YouTube, creators, Reddit, technical communities, and peer tools. Primary KPIs: authorization, install, activation, workspace usage, and retention. This is where AI Mode’s 23% video share matters most, because the primary AI destinations are OAuth, MCP, and workspace URLs that ordinary editorial content will not surface.

For Accounting and Finance, the job is “reduce risk and make financial decisions.” Owned priorities: guides, calculators, comparisons, compliance explainers, and support. External priorities: specialist finance publishers, peer vendors, reviews, and communities. Primary KPIs: signup, filing, calculator use, support completion, and account access. Finance also has the smallest citation-to-traffic gap, which means guides earn both the citation and the click more often than in the other two categories. Our GEO agency buyer guide covers how to select a partner who understands these category-specific splits, and the GEO content audit workflow shows the four-pass framework that catches missing owned assets and missing external corroboration in one pass.

How to Act on Saas AI Search Optimization This Quarter

You do not need to rebuild your whole content strategy this month. You need to add three workstreams to what you already run, and update your reporting so the split shows up in your dashboards.

Workstream 1: Fund the external ecosystem as a first-class channel. With 84% to 93% of citation weight sitting on third-party sites, this is not a PR afterthought. Give it an owner. Fund review-site accuracy, community presence on Reddit and category forums, peer-vendor documentation and integration listings, and target publisher mentions. The Sitecore Scrunch acquisition and the broader Answer Engine Optimization platform wave are direct evidence that enterprise tooling is catching up to this shift.

Workstream 2: Split distribution by platform. ChatGPT gets structured written content, reusable evaluative pages, and canonical brand surfaces. AI Mode gets video, creator content, community and social presence, and deep task-oriented URLs. If you only fund one, you concede half the AI landscape. Reddit is now a real citation source, not just a social channel, which is why our team wrote the Reddit dual attribution operator playbook to help teams measure Reddit’s true contribution.

Workstream 3: Report citations and referrals separately. Update your GA4 setup so citation-driven visits, AI referral visits, and organic search visits are three separate channels with their own conversion metrics. Our GA4 audit playbook covers the consent and channel-tagging changes you need in place before you can trust the split. Then define destination quality per category, so a homepage visit in CRM and an OAuth redirect in Collaboration are not evaluated against the same benchmark.

The brands that will win the next 18 months of saas ai search optimization are the ones that stop treating AI visibility as a content problem and start treating it as a three-layer operational program: owned assets, external ecosystem, and measured destinations. If you want a second set of eyes on your current split across those three layers, our team runs the same framework on client accounts. Book a free consultation and we will show you where your citation and referral share sits against category norms.

Let’s Grow!