The main KPIs for AI search are 3 numbers: LLM traffic, LLM conversions, and the revenue or pipeline behind them. All 3 live in your analytics platform and your CRM, not in your AEO tool. Everything the AEO tool reports, AI visibility included, is a signal: a leading indicator that tells you whether those 3 numbers are about to move.
The first version of this article argued that visibility isn't the KPI, then skipped the obvious follow-up: fine, then what is? This version answers it.
What are the main KPIs for AI search?
LLM traffic. Visitors arriving from AI assistants: referrals from chatgpt.com, perplexity.ai, claude.ai, gemini.google.com. You find it in GA4 or whatever analytics you run, by filtering on referrer.
LLM conversion. The share of those visitors who become signups, leads, or demos. Same analytics platform, plus your CRM when the sale closes offline.
AI-sourced revenue or pipeline. What those conversions are worth. This is the number a client renews a retainer on.
2 things about these numbers:
- They lag. By the time LLM traffic moves, the content work that caused it happened months earlier.
- They undercount. Plenty of people read an AI answer, then google your brand or type the URL directly. Analytics logs that as organic or direct. The real effect of AI search is bigger than what the referrer data shows.
Which is exactly why you can't manage AI search on KPIs alone. You need the layer that moves first.
Leading vs lagging KPIs in AI search
A lagging KPI measures the outcome after it happened. Revenue this quarter. You can't push on it directly, you can only confirm it moved. A leading indicator measures what predicts the outcome. It moves earlier and it's noisier. It's also the only thing you can act on this week.
Sales teams have run on this split forever: calls booked is leading, deals closed is lagging. AI search has the same 2 layers.
Lagging (the KPIs): LLM traffic, LLM conversions, revenue.
Leading (the signals): AI visibility, mention rate, position, sentiment, source citations. Everything in the metrics worth tracking.
!The AI search measurement chain: content work feeds leading signals like AI visibility and mention rate, which feed lagging KPIs like LLM traffic, conversions, and revenue
The chain runs in 1 direction. Content ships. Models start citing it, and visibility moves within days or weeks. Humans see the brand recommended, click through or search for it, and the KPIs move over months. A visibility number that rises today is next quarter's traffic, if the content behind it holds up.
Why AI visibility is volatile (and why that's fine)
Ask ChatGPT the same question about your brand twice and you can get 2 different answers. Retrieval changes, training data refreshes, sampling adds randomness. That's not your tracker breaking, it's the nature of the thing being measured.
It's also normal for a leading indicator. Leading indicators are noisy in every field, that's the price of moving early. The mistake is treating a single volatile reading as a verdict. A brand that jumps from 40% to 55% mention rate in a week might have improved, or might have had a lucky sampling week. Read signals as trends over 4 to 8 weeks. Judge KPIs monthly or quarterly, on a timeline long enough for the noise to average out.
> Signals can be volatile. What matters isn't the number, it's the action you put behind it.
You get 50% AI visibility? So what? What does that change in your next iteration? If the answer is nothing, ignore that signal and focus on the one that helps you make better content for the people asking.
SEO fell into this trap with rankings and built a decade of dashboards on chasing a noisy number. AEO doesn't have to repeat it. Don't SEO the GEO.
How AI visibility improves AI acquisition
The signal isn't the goal. It earns its place another way: it's the only layer that tells you what to fix.
Here is what that looked like on my own brand. When I started tracking AEO Copilot with its own tool on 27 August 2026, the baseline was 1 mention across 28 answers (7 prompts, 4 engines). That 1 mention came from Perplexity, at position 6, for "best AEO tools for AI-native startups and solo founders", and the page it cited was my own LinkedIn article, not the website. A lagging KPI would have told me "no AI traffic," which I already knew. The visibility data said something more useful: nothing on aeo-copilot.com was worth citing for that question. That's not a score to report. That's a to-do list.
So I worked the list. On 28 August I reworked the 5 posts aimed at AI-native startups. On 1 September, the same prompt put AEO Copilot at position 1 on Perplexity, and this time the cited page was the best AEO tools guide on the site. I can't prove the rewrite caused it; 1 engine and 1 prompt is a small sample. But it is the pattern the loop below predicts.
It also shows why the signal is not the goal. In the week of 27 September, across 30 tracked prompts, AEO Copilot had 3 mentions, all on Perplexity, and still 0 on ChatGPT, Claude and Google AI Overviews. The signal moved on 1 engine. Whether LLM traffic follows is the KPI question, and that answer comes weeks later.
!The AI acquisition loop: track visibility weekly, find the gap, fix the specific content, get cited, and LLM acquisition follows
- Track visibility weekly, across platforms and prompts.
- Find the gap. Which prompts you lose, which platform is weak, which page never gets cited. Usually it's 1 of these reasons your brand isn't showing in AI answers.
- Fix the specific content. A pricing prompt returning vague answers means the pricing page doesn't answer the question.
- Get cited. The models pick the page up (the mechanics: why your brand isn't showing up in AI answers). Mention rate and position move first.
- Acquisition follows. LLM traffic and conversions move weeks or months later.
Then repeat. Without the signal layer, the loop breaks at step 2: you know acquisition is flat, but not why or where. That's how AI visibility improves AI acquisition. It doesn't work as a number to maximize. It works by pointing every content sprint at the exact prompt and page where you're losing. For the technical half of step 2 (robots.txt, schema, llms.txt), run a free AEO audit, or check a single page's LLM readiness with the website audit.
How to set up AEO KPIs for your clients
Pick the KPI with the client first, tied to revenue or pipeline, before anyone opens a visibility chart. Agree on what counts as an AI-sourced lead and where it gets measured. That 1 conversation prevents every "but the score went down" call later.
I never present a lone visibility percentage and call it the report. The report leads with the KPIs: what happened to LLM traffic, leads, and pipeline this period. The signals come after, as the explanation and the plan. Visibility rose on comparison prompts after the rewrite shipped, so next month's traffic should follow. Sentiment dipped on pricing prompts, so that page is next.
If visibility moved and nothing downstream moved for months, that's a flag to dig, not a win to celebrate. If visibility barely moved but 1 high-intent page started pulling qualified AI referrals, that's the actual story.
The delivery side of this has its own rule: don't explain AEO to your clients, show them. And if you're packaging AI visibility as a recurring service, start with the AEO guide for freelancers and agencies.
Frequently asked questions
What are the main KPIs for AI search?
LLM traffic (visitors arriving from AI assistants), LLM conversion (the share of those visitors who become leads or customers), and the revenue or pipeline behind them. All 3 live in your analytics platform and CRM. AEO tool metrics like visibility and mention rate are leading indicators, not KPIs.
What's the difference between a leading and a lagging KPI in AEO?
Lagging numbers (traffic, conversions, revenue) confirm an outcome after it happened. Leading numbers (visibility, mention rate, position, sentiment) move earlier and predict it. You act on leading numbers weekly and judge lagging numbers monthly or quarterly.
Is AI visibility a useless metric?
No. It's the layer that moves first, and the only one that tells you where to focus: which prompts you lose and which pages never get cited. The mistake isn't tracking it, it's reporting it as the goal.
Why does AI visibility change from one check to the next?
The models are non-deterministic. The same prompt at different times can pull different retrieval results or sampling and return a different answer. That's expected behavior, and the reason to read visibility as a 4-to-8-week trend instead of a single reading.
What should I report to a client or stakeholder?
Lead with the KPIs: what changed in LLM traffic, leads, and revenue. Use the signals as the evidence for why it changed and the plan for what to fix next. A signal-only report ages badly the first time the number wobbles.
Summary
- The main KPIs for AI search are LLM traffic, LLM conversions, and AI-sourced revenue. They live in analytics and CRM, not in your AEO tool.
- KPIs lag: they confirm outcomes months after the work. Signals lead: AI visibility, mention rate, position, sentiment, and citations move within weeks and predict the KPIs.
- Signals are volatile because the models are non-deterministic. Read them as 4-to-8-week trends, never as single readings.
- AI visibility improves AI acquisition by locating the fix: the lost prompt, the weak platform, the uncited page. Track, find the gap, fix, get cited, repeat.
- Report KPIs first, signals as the explanation. Don't SEO the GEO.
For the full list of signals and where each one lives: AEO metrics that actually matter (and where to find them). For the monitoring setup: How to track brand mentions in ChatGPT and Claude.
AEO Copilot tracks the signal layer, visibility, mention rate, position, sentiment, and citations, across ChatGPT, Claude, Perplexity, and Google AI Overview. The free plan gives you 1 brand and 50 prompts to find your baseline.