Last Updated on October 29, 2026
This was a question raised in the last Podcast AMA I did together with Rob Cairns. It is a good question, because it is generally where traditional SEO reporting starts to fall short.
My answer? Keep rankings, add structured AI visibility checks, and measure business results separately.
Rankings are not visibility
Picture someone in your field asking ChatGPT for a recommendation. The answer draws on expertise like yours, maybe names your brand, and the person never opens Google. Would your ranking report show that?
The answer is no.
We used to measure visibility with rankings, impressions, clicks and traffic. Someone searched, saw us in the results, and hopefully clicked. Simple enough. But Google now answers many questions directly, and AI tools answer without Google at all. The answer may include our expertise, our content or our brand, and nobody clicks. That is zero-click search, and a ranking report tells us very little about it.
We need to separate rankings from visibility.
What still works, and where it falls short
Traditional search metrics still matter; they just describe only one kind of encounter, where someone saw a results page.
Still useful:
- Rankings show where a page appears for a query.
- Impressions show how often it appeared.
- Clicks and organic traffic show who came through.
Where they fall short:
- A person reads an AI answer that uses your content, never visits your site, and nothing is recorded.
- Google answers directly on the results page and sends no click.
- A strong ranking can sit next to a weak click-through rate and the report cannot explain why.
I am not asking you to throw away your SEO report. I am asking you to stop treating it as the whole picture. Ranking is still a useful metric. It just isn’t the whole story.
In fact, SEO isn’t the whole story. Honestly, how many centuries did businesses successfully market to customers before SEO?
So what would I report instead? Three things: Search performance. AI visibility. and Business results.
The three-layer report
Report three things and keep them separate, so one layer’s weakness cannot hide behind another’s strength.
| Layer | Question it answers | Metrics |
|---|---|---|
| 1. Search performance | Where do our pages appear, and who clicks? | Rankings, impressions, clicks, organic traffic |
| 2. AI visibility | Do AI answers mention or cite us, and how consistently? | Mentions, citations, rate across repeated queries |
| 3. Business results | Does any of this turn into something useful? | Branded searches, enquiries, referrals, conversions |
To make it concrete: my website warrenlainenaida.net. It has English and German versions, SEO and AI books, workshops, a monthly Digital Edge package and a Smart Seniors hub.
Layer 1: Search performance (keep as is)
Nothing changes here. Report the familiar metrics from Search Console, as you always have.
| Query | Avg. position | Impressions | Clicks | CTR |
|---|---|---|---|---|
| seo all you need to know | 1.4 | 1,900 | 610 | 32% |
| seo book for beginners | 7.8 | 6,400 | 190 | 3.0% |
| seo training for nonprofits | 11.2 | 1,100 | 22 | 2.0% |
| SEO Buch für Anfänger | 5.1 | 2,300 | 140 | 6.1% |
Look at the second row. Position 7.8, lots of impressions, 3% click-through. A ranking report cannot tell me whether that is normal or whether an AI answer is absorbing the click. That is my cue to look at layer 2.
Layer 2: AI visibility (test repeatedly)
This layer asks whether your brand or content appears in AI answers, whether it is cited, and how consistently that happens across repeated queries.
20 prompts in four clusters of five, written in English and German because the site has both. Run them weekly on ChatGPT, Gemini and Perplexity, three runs each. That is 180 observations a week, so a tool or a disciplined sample does the work, not a person pasting answers by hand.
Log every observation in the same shape:
| Date | Platform | Prompt | Mentioned? | Cited URL | Competitors named |
|---|---|---|---|---|---|
| 12 Oct | ChatGPT | “Good beginner book on SEO and AI?” | Yes | /ai-seo-a-new-book/ | 2 |
Then report rates by cluster, not single results:
| Prompt cluster | Mentioned | Cited with link |
|---|---|---|
| Books (“beginner SEO book”) | 58% | 31% |
| Seniors’ digital skills | 34% | 20% |
| SEO training and workshops | 22% | 9% |
| Marketing help for a nonprofit or school | 8% | 3% |
The brand shows up where the books are and almost nowhere where the consulting revenue is. A ranking report would never have shown me that gap. Notice also that mentioned and cited are different columns. Being named without a link is still exposure, but it sends no traffic.
Layer 3: Business results (does it matter?)
This is where you check whether visibility turns into anything. Report it on its own, next to layers 1 and 2, and do not claim that one caused the other.
| Metric | September | October |
|---|---|---|
| Branded searches (“Warren Laine-Naida”) | 140 | 185 |
| Referral sessions from chatgpt.com and perplexity.ai | 11 | 19 |
| Contact form submissions | 9 | 12 |
| Free-call bookings | 4 | 6 |
| Digital Edge package enquiries | 1 | 2 |
| “How did you hear about us?” answered “AI assistant” | not asked | 3 |
The last row is the important one. The current contact form asks for email, message, consent and phone, so the practical step is to add a “How did you hear about us?” dropdown with an AI assistant option.
Three people told me they found me through an AI assistant. Analytics would have filed all three as direct or branded traffic. Without that one form field, I would have seen the branded-search bump and never known why.
The catch: treat AI citations as probabilities
One AI citation does not mean lasting visibility. Results change with the wording of the question, the platform, the date and the context.
- One citation is an anecdote, not a ranking.
- Test repeatedly, vary the phrasing, and record what you actually observe: query, platform, date, cited or not.
- Report a rate (“cited in 6 of 10 runs”), not a position.
This is the biggest mindset change from rank tracking. With rankings, one number was the answer. With AI answers, the honest answer is a rate, and a rate comes with uncertainty. I would rather show you ‘somewhere between 30% and 90%’ than a confident 60% built on ten runs.
The query-count section near the end shows the margin of error for 10, 30 and 100 runs, if someone asks.
Be honest about what we cannot measure
Some of the value of AI visibility will never appear in analytics, and a good report says so.
Someone reads an AI answer, remembers your name, and contacts you three weeks later. Unless they tell you how they found you, that connection may never appear in your data. Analytics will call it direct or branded traffic.
- Add a “How did you hear about us?” question to enquiry forms, with AI assistants as an option.
- Watch branded search trends as a proxy for people who met your name elsewhere.
- State the gap plainly in every report, so nobody reads a flat line as proof of no effect.
Wrapping Up
Rankings tell us where a page appears. Visibility tells us whether people encounter our expertise. Business results tell us whether that exposure matters.
So my approach is simple. Keep traditional SEO reporting, add structured AI visibility checks, and measure business outcomes separately. If most answers never reach a traditional results page, ranking is still a useful metric. It just isn’t the whole story.
Add the “How did you hear about us?” field to your enquiry form this week. It costs nothing and starts building the evidence layer 3 depends on.
My Motto atm
- We’re measuring the wrong things.
- Watch what people do, not just what you can measure.
- Just because we can’t see an impact doesn’t mean it isn’t there.
- Tech changes how people behave, and our metrics and methods must change with it.
Two Extra Questions
Which tools track AI visibility?
Start with a spreadsheet and GA4, and add a paid tool only when manual logging becomes the bottleneck. The tool market is crowded, and the comparisons below come from vendor roundups seen as search results on 26 October 2026, so treat features and prices as claims to verify.
| Option | Best for | Note |
|---|---|---|
| Spreadsheet plus GA4 | Small teams, a first baseline | One guide suggests GA4 referral segments for chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com (SEOcrawl) |
| Otterly.AI | First paid or free-tier test | Reported to have a free tier and to cover ChatGPT, Perplexity and Google AI Overviews (Intempt); another roundup lists plans from $29 a month (Alhena) |
| Semrush AI visibility add-on | Teams already on Semrush | Sold as an add-on to the existing SEO platform (Rivo) |
| SE Ranking AI Results Tracker | Teams already on SE Ranking | Tracks mentions and linked citations across AI Overviews, AI Mode, ChatGPT, Gemini and Perplexity (INSIDEA) |
| Peec AI | Wide engine coverage | Listed as covering ten AI engines (INSIDEA) |
| Profound | Enterprise analytics | Positioned as an enterprise option (metehan.ai) |
Whatever you pick, check which platforms it covers. Visibility in ChatGPT can look quite different from visibility in Perplexity or Google’s AI Overviews, so a tool that watches only one gives you a partial picture.
How many queries are enough?
Start with 20 to 30 prompts tied to your core topics, run each 3 to 5 times per platform, and keep the set stable. This is a rule of thumb, not an industry standard.
- Prompts: group them in clusters of 5 to 8 by topic, in each language you serve. Add new prompts rather than swapping old ones, so trends stay comparable.
- Repeats: a single run is an anecdote. Repeating each prompt is what turns a citation into a rate.
- Size of the effort: 20 prompts on 3 platforms with 3 runs each is 180 observations per round.
The reason to repeat is the margin of error. For an observed 60% citation rate, the rough 95% margin looks like this:
| Runs behind the 60% | Rough margin | Honest way to report it |
|---|---|---|
| 10 | about ±30 points | between 30% and 90% |
| 30 | about ±17 points | between 43% and 77% |
| 100 | about ±10 points | between 50% and 70% |
Report ranges, and treat a month-to-month change as real only if it clearly exceeds that noise. A move from 55% to 60% on thirty runs is not news.