Rethinking content KPIs in the AI search era

The KPIs most B2B tech content teams reported on through 2024 do not survive contact with AI search. Sessions, pageviews and time-on-page were imperfect even before LLMs entered the picture. Now they’re actively misleading. A site can lose half its organic sessions while doubling its commercial impact, because the buyers are still being influenced, just not via clicks that GA4 can see.

This piece walks through the KPIs we’d retire, the ones we’d keep and the ones we’d add. It’s based on what we’ve seen work for clients across MSPs, SaaS and ERP consultancies through 2025 and into 2026.

What’s broken with the old set

The classic content KPI set looked something like organic sessions, organic users, time on page, pages per session, scroll depth, conversions and assisted conversions. Each of those still has a use, but each of them is now distorted by AI search in ways that need acknowledging.

• Organic sessions is shrinking on many sites, even where commercial influence is steady or growing, because Google AI Overviews and ChatGPT are answering some queries without sending clicks.
• Time on page is rising slightly on average for the sessions that do arrive, because AI-referred traffic is more pre-qualified. That sounds good. It’s actually noise, because the underlying volume is changing.
• Conversions still happen, but the path to them increasingly starts in an LLM that GA4 cannot see. Last-touch attribution makes the AI surface look insignificant when in fact it was decisive.

We covered the attribution-shaped half of this in tracking AI search traffic and the wider attribution problem in attribution models for multi-touch tech buyers.

What to keep

Some KPIs still pull weight, with caveats:

• Branded search volume. When LLMs surface your brand in answers, branded search rises, often weeks before any other metric moves. Watch this in Search Console and any brand-tracking tooling you use. We’ve split the two sides of this measurement out in branded versus unbranded AI search.
• Direct traffic to deep pages. A spike in direct traffic landing on a specific case study or comparison page is usually AI referral with the referrer stripped. Track the shape, not just the volume.
• Conversion rate from AI-referred sessions. Where you can identify them, AI-referred sessions tend to convert above average. Worth segmenting.
• Demo, contact and trial actions. The bottom-of-funnel actions still count and are still measurable.

The shift is from volume metrics towards quality and qualification metrics. A smaller number of higher-intent sessions is better than a larger number of casual ones, and the data should reflect that.

What to retire (or relegate)

A few metrics we’d argue are now harmful to over-index on:

• Pageviews and sessions as a primary KPI. They tell you less about commercial influence than they used to. Useful as a directional input, dangerous as a target.
• Bounce rate. Always weak, now genuinely meaningless when AI traffic lands deep, gets the answer it needs and leaves satisfied.
• Pages per session. Same logic. A high-intent visitor reading one page and booking a demo is a better outcome than three pages and no action.
• Generic “engagement” composites. GA4’s engagement metric muddles signals. Treat it as a smoke alarm, not a thermostat.

What to add

The new metrics that earn their place:

Citation share of voice

For your priority topics, what percentage of LLM answers cite you versus your competitor set? Tools like Profound, Athena and Semrush’s AI tracking will give you this. Manual audits supplement.

This is the single best leading indicator we’ve found for AI search visibility. It moves before referral traffic does, and it tracks something that pageviews never could: whether your brand is being mentioned in answers, regardless of click-through.

Prompt-level visibility

Within citation share of voice, drill down to specific prompts. The “best MSP for law firms in Birmingham” prompt is a different battle to “what is co-managed IT”. Different content needs to win each. Tracking at prompt level forces the team to work the right pages, not just the easy ones.

Crawl coverage by LLM bots

From server logs or Cloudflare. Are GPTBot, ClaudeBot, PerplexityBot and Google-Extended crawling your priority pages at a useful frequency? If not, the ceiling on AI search visibility is structural, not creative.

Brand mention volume in third-party content

Backlinks still matter. Mentions without links matter more than they used to. Tools like Ahrefs, Semrush and Google Alerts can track this. We covered the underlying logic in brand mentions vs backlinks in AI search.

Pipeline-influenced rate

The hardest one, and the most important. Of the deals that closed this quarter, how many touched content during the buying journey? CRM-side, this is messy work involving form fields (“how did you hear about us”), surveys, sales-rep tagging and sometimes deal-coding by hand. It will never be perfect. Done roughly, it’s still better than guessing.

For B2B tech with long buying cycles, this is the metric that actually matters. We’ve covered the operational side in measuring content marketing ROI for tech.

A simple KPI table for 2026

Here’s a structure we use with clients as a starting point.

The point is to separate outcomes (pipeline) from leading indicators (citation and crawl) from activity (what the team did). Most content reports we see are activity-heavy. The reports that justify investment are the ones that show a defensible chain from activity to leading indicator to outcome.

Reporting cadence and audience

A practical rhythm we’ve used with B2B tech marketing leads:

• Weekly internal stand-up. Activity and leading indicators only.
• Monthly leadership update. Leading indicators and outcomes, with prompt-level visibility called out for the priority topics.
• Quarterly board pack. Outcomes first, with a single-page summary of leading-indicator trends and the qualitative read on what’s changed.

The biggest mistake we see is reporting too much detail too often. Citation visibility data fluctuates week to week in ways that are mostly noise. Reporting it weekly invites overreaction. Reporting it monthly with a sensible commentary lets the signal emerge.

A final note on incentives

Whatever KPIs you pick, the team has to be incentivised on them. We’ve watched marketing teams adopt new AI-era KPIs in their reporting while still being reviewed against the old volume targets. That does not work. If pageviews are flat or down because answers are being cited instead, the team needs to be measured on something other than pageviews. This is a leadership conversation, not a tooling one.

If you’d like a second opinion on your AI search strategy, drop us a line. KPI redesign is one of the more useful conversations we have with new content marketing and AI SEO clients, and it’s usually overdue.