The real story behind the 53% drop in SaaS AI traffic

As the SaaS market moves away from sales driven by independent AI agents like Claude Cowork, new data shows a 53% drop in AI-driven discovery sessions. Wall Street has dubbed it the “SaaSpocalypse.”

Whether AI agents will replace SaaS products is a bigger question than that dataset can answer. But panic is already distorting the interpretation, and this data cuts through the noise to show what SEO teams should really be watching.

Pilot went from 0.3% to 9.6% of SaaS AI traffic in 14 months

From November 2024 to December 2025, SaaS sites logged 774,331 LLM sessions. ChatGPT drove 82.3% of that traffic, but Copilot’s growth tells a different story:

SaaS AI Traffic by Source (Nov 2024 – Dec 2025)

Starting with only 148 sessions at the end of 2024, Copilot grew more than 20x by May 2025. From May to December, it reached 3,822 times per month, making it the second largest AI referring SaaS sites by the end of 2025.

Investors have written off $300 billion in SaaS markets over fears that AI agents will replace enterprise software. But this data points to a much smaller force: proximity.

Copilot thrives because it captures intent within the workflow. Standalone tools saw a 53% decrease in traffic while AI embedded in the service grew by 20x.

Testing software is work, and Copilot lives where that work happens.

When someone asks, “Which CRM should we use for a group of 20 salespeople?” when creating a business case in Excel, that time is taken – one ChatGPT does not see. May’s move reflects that opening: Microsoft 365 users realize they can research software without opening a new tab.

41.4% of SaaS AI traffic comes from internal search pages

SaaS AI discovery sends users to internal search results first, not product pages.

Top SaaS landing pages by LLM volume

Despite capturing 320,615 times – more than the blog, prices, and product pages combined – this dominance probably reflects LLM’s limitations, not superior content. LLM users move users to search when they don’t have a specific answer.

For SaaS companies watching their stock crater, that’s good news: there are concrete technical fixes. 41.4% is not an existential threat. Clarity is a problem.

If LLM can’t find a specific answer, it automatically switches to the site’s internal search. AI treats your search bar as a trusted backup, assuming that the search schema will produce the right page even if a product page is not in the index.

At 1.22%, the search page penetration is 8.7x the site average. The reason is a “safety net” effect, not a full optimization.

If more specific pages – like Product or Pricing – don’t have the data LLM needs, it goes back to the broader search results. LLMs see the structure of a search URL and hope it returns something important, even if they can’t predict what.

Blog pages followed with 127,291 sessions and 1.13% traffic. This is an organized comparative post – “the best CRM for small teams” or “Salesforce alternatives” – LLMs cite where they have specific recommendations.

Pricing pages show a 0.45% entry; product pages, 0.28%. When users ask about software options, LLMs go to comparison sites — search and blog — first. Specific product pages or prices are only quoted if the inquiry is specific to the seller.

July’s high and Q4’s low reflect business cycles

SaaS AI traffic increased in July by 146,512 times, then slightly decreased in Q4:

The whole field declined. ChatGPT volume was cut in half, falling from 127,510 sessions in July to 56,786 at the end of the year. The pilot decreased from 4,737 to 2,351. Perplexity decreased from 7,475 to 3,752.

Two factors drove this slide:

• People were not working. August is holiday season, November includes Thanksgiving, and December is the holidays. Software research takes place during work; when offices close, adoption goes down.
• Q4 ends the financial “buying window”. Many teams have spent their budget for the year or postponed contracts until Q1 funding opens. Even the teams that are still working are not testing the equipment because there is no budget left until the new financial year.

July’s peak reflects mid-year momentum: people are working, and the Q3 budget is still available. The decline in Q4 reflects both fewer researchers and fewer active buying cycles.

This is where the sales narrative breaks down.

Investors treat the 53% drop in traffic as evidence that AI adoption has stalled. But the data is consistent with typical B2B financial cycles.

AI does not fail as a channel for information acquisition. It falls into the same seasonal rhythm as all other B2B shopping methods.

What does this data mean for SEO teams

Raw traffic numbers do not indicate where to invest. Browsing rates and landing page distributions reveal what matters.

Track traffic by page type, not global averages

SaaS shows AI penetration of 0.41% across the board, but that ratio masks the focus. Search pages reach 1.22%—8.7x more. Blog pages reached 1.13%. Price pages are 0.45%. Product pages remain at 0.28%.

If you’re only tracking total AI sessions, you’re measuring the wrong metric. AI traffic can increase by 50% while landing on high value pages decreases. Volume hides what matters: where AI users focus when they arrive with intent.

Action:

• AI traffic segment by page type in GA4 or your analytics platform.
• Track traffic (AI sessions ÷ total sessions) by page category every month.
• Identify pages with high focus, and optimize those areas first.

Search results pages are now the first place of discovery

Internal search accounts for 41.4% of SaaS AI traffic. If those results can’t be clear, identifiable, or structured for comparison, you’re not visible to the larger segment of AI-driven consumers.

Most SaaS sites treat internal search as navigation, not content. The results return grouped listings with minimal product details, no filter signals in URLs, and content rendered in JavaScript that LLMs cannot parse.

Action:

• With 41.4% of traffic hitting internal searches, treat your search bar as an API for AI agents.
• Make search pages clear (check robots.txt and indexability).
• Add structured data using SoftwareApplication or product schema.
• Local comparison data — prices, key features, number of users — directly results, not just product names.

Make your data relevant to LLMs – both values ​​and content

Pricing is outdated, but for many SaaS companies the real risk is invisible. Pricing pages show AI penetration of 0.45%—below the cross-industry average of 0.46%. Blog pages captured 127,291 times for 1.13% of traffic, but only when the content directly answered the opt-in questions. The pattern is clear: LLMs say what they can learn and analyze. Skip those who can’t.

Many SaaS sites still include pricing on the back of contact forms. If the price requires a sales discussion, AI will not recommend “tools under $100/month” questions. The same applies to blog content. When someone asks, “Which CRM should I use?” LLM looks at posts that compare options, explain decision making, and explain trade-offs. General thought leadership in CRM trends is not presented.

Action:

• Publish prices on a dedicated, transparent page. Include representative examples, seat minimums, contract terms, and exclusions.
• Keep prices transparent. Explicit pages are cited; portal pages do not.
• Replace standard blog posts with structured comparison pages. Use tables and delete data points.
• Remove the fluff. Provide basic data that allows AI to verify compliance and integration capabilities in seconds, not minutes.

AI embedded in the workplace grows 10x faster than standalone LLMs

Copilot grew 15.89x year over year. Claude grew by 7.79x. ChatGPT grew by 1.42x. Rapid growth in tools embedded in existing workflows.

Workplace AI is changing the context of discovery. In ChatGPT, users search transparently. In Copilot, they ask questions that are central to the task—creating a proposal, creating a comparison spreadsheet, or reviewing vendor options with their team.

Action:

• Follow Copilot and Claude separately transferred to ChatGPT. Note which pages these sources like.
• See the intent: these users aren’t browsing — they’re in the middle of a task, deep in exploration, and close to a purchase decision.
• Expose AI discovery to work to support real-time purchase adjustments.

Survival loves what is available

A 53% drop from July to December shows the use of AI is stabilizing in the software procurement process. Consumers are learning which decisions benefit from AI integration and which do not. The traffic is very deliberate, focused on complex evaluations where comparison is important.

For SaaS companies, the early positioning window is closing. The $300 billion in retail sales affects the industry broadly, but the companies that survive the refinance will be the ones consumers can find when they ask an AI agent, “Should we renew this contract?”

Teams investing now in transparent pricing, transparent data, and benchmark-focused content are building that presence while competitors debate whether AI adoption is worth it.