The Metrics We Built Are Breaking. SEO Success Needs To Look Different Now. 

There’s a version of this quarter’s report where organic traffic is flat or slightly down. The room gets uncomfortable. You spend twenty minutes explaining why the number your board is looking at doesn’t reflect the actual impact of the work you’ve been doing. You’re quietly sweating because you feel it’s the only lever you have to justify your salary. They’re questioning your value, and their trust in your expertise. 

It’s not that the work isn’t valuable or important or necessary for a business in the age of the Internet. It’s that the metrics we’ve been using by default to prove it are no longer up to the job. 

It’s not sessions, or clicks, impressions or even revenue.

In fact, those never were the right metrics, because as an industry we’ve fundamentally misunderstood — deliberately or through lack of awareness — what search engine optimisation is as a channel. 

We are marketers. Full stop. 

I don’t have a clean, finished framework. Anyone who says they do is lying…except maybe Jono (but we’ll talk about his position later). What I do have is a set of replacements I’ve been testing, some borrowed from marketing disciplines that have dealt with attribution problems for far longer than we have.

What Stopped Working

A few things broke at roughly the same time. It broke a lot of our brains, and some of our jobs. 

AI Overviews and LLMs are eating informational clicks visibly and quickly 

Pew Research tracked nearly 69,000 real searches, and click rates on traditional links dropped to 8% when a summary appeared, down from 15% without one. Ahrefs put the hit at 58% for position one by February 2026.

Yet Semrush tracked the same keywords before and after they got AI Overviews and found zero-click rates didn’t actually go up. Those were already queries that weren’t driving clicks, before AI came in and made it obvious. AI Overviews didn’t take traffic away. It showed up where most folks weren’t clicking already anyway.

A lot of what we’ve been reporting as “organic growth” (yes, even incremental, year-on-year growth) has been clicks to informational queries padding the graph without moving the bottom line, and more importantly, likely without much brand recognition, either — because the content was probably boilerplate, FAQ style, perfunctory language articles that were “done for SEO.” 

Years of an obsessed and frantic chase to rank for arbitrary high-volume search terms as a representative measure of success in organic channels has left its mark on the discipline. We stopped looking for visibility in topics that mattered, and instead went for topics that drove volume. 

Because for a long time, that worked. 

We could live in that gap. Visible, but not…superior. Visible, but not a brand. exactmatchdomain.com to the nth degree. Brand trust wasn’t necessary. Brand sentiment was disregarded. Brand recognition slowed us down. 

But change is the only constant, and the maturity of folks on the Internet shifts quicker than we catch it, a lot of the time. AI has accelerated that.  

So, AI Overviews and LLM’s didn’t break something that was actually working. The technology exposed the ouroboros of measurement bullshit of //up and to the right// we created ourselves by chasing head terms rather than a good business and a good product.

The measurement model we’ve been using for organic visibility and impact has never actually been the right one, and a lot of us knew it.

Don’t lie. We’ve all had that sinking feeling when we’ve actually looked at page-level organic sessions uplift and it’s been a glossary page or a 300 word definition with a 2-second page dwell. 

And most of us — particularly if we worked agency-side at some point in our career — have reported it as growth anyway because we needed to put food on the table. It was, and still is, the status quo. When you don’t look too closely, it works. 

We have no choice but to look more closely now.

Attribution across Google’s own surfaces is a mess.

As an early canary, GA4 itself stripped out what teams relied on. Page value disappeared, events replaced sessions, and generally, the curtain was pulled back and the data interpretation actually felt difficult when it hadn’t before. All before any of this AI stuff even arrived. Reporting itself got more difficult before large language models and poor multi-surface tracking made it worse. 

And then it all got more confusing.

Discover, AI Mode, AI Overviews land most often in GA4 as (direct). The “Bermuda triangle of measurement”:

• Search Console lumps AI impressions in with regular search
• Bing folds Copilot data into standard web metrics
• Lily Ray showed this concretely: 363K Discover clicks on one article, practically no traffic reported in GA4

The visibility exists. Unified, clear measurement doesn’t.

LLM visibility and its impact is growing but mostly invisible. 

93% of AI Mode searches don’t produce a click and it’s up significantly year on year (though please remember the law of small numbers, we’re still living in the framework with AI as a channel). The visits that do come through convert at 4.4× standard organic with much lower bounce — a high-quality stream that sessions-based reporting treats as noise. 

This is alongside the fact that for every 1,500 times ChatGPT pings your server, they generally only send one referral. One visit. Signal and noise and we can’t tell which is which.

These three situations all point back to one problem: organic search interest and visibility is fractured across surfaces and platforms our standard reporting was never built to see or capture in a tangible way.

Where the New (Old) Metrics Come From

Digital growth as a whole is heading to directional measurement. More estimation, more educated guesses, less certainty about which action caused which outcome. For many reasons, only one of which is the impact of LLM’s and AI. (Think privacy laws, JS-blocking, app-to-web and dark social, among many others). 

That feels like a loss if you’re used to last-click precision. But that precision was always overstated. Simplified to the point of near-irrelevance. 

We can’t remove AI visibility from our reporting now. Pandora’s box is opened. But we can stop trying to force it into the box of old digital metrics like traffic, and wrap it into the larger conversation around visibility and impact we need to be having.

My first piece of advice has been consistent for years: go talk to finance. Ask how the business measures TV, radio, sponsorship. Channels that have never had clean attribution. If you can’t get to finance, try your brand team. They’ve been working with approximations for decades. Businesses still invest because directional measurement is better than none.

And if you have no other recourse, set the value of the goal you’re aiming for to $1. Start the conversation. 

Anchor One: Share Of Voice

Binet and Field established decades ago that brands whose share of voice exceeds their market share tend to grow. Theirs was a TV-and-print rule. The translation to organic isn’t airtight — nobody’s run that analysis with the rigour the IPA brought to media spend — but the logic holds. If more people are searching for you and your category competitors than for the rest of the market, you’re growing the salience of your brand in the place buying decisions happen.

This is also where most “AI visibility” tools are quietly trying to plant a flag. Just badly. Same idea, opaque execution. We’ll get to that.

If you don’t have the budget, the cleanest implementation right now is share of search via Google Trends and a watch on share of category visibility in your top 20 SERPs. If you do have the cash to float on what essentially amounts to an experiment, choose a platform like Demandsphere or Profound and stick to it. Know how it works, never report on individual query movement, and use it representationally.

Pick one and stick to it. Doesn’t matter much which. Consistency matters more than the choice.

Anchor Two: Branded Search Trend

If your work is actually building anything — recognition, salience, the next-purchase consideration set — branded search should grow. Slowly, lumpily maybe, but it should grow. It’s the simplest leading indicator we have, available in Search Console for free, and your stakeholders already understand what it is without you explaining it.

If branded search is flat across a year of sustained content and PR work, something isn’t landing. That’s worth knowing.

Anchor Three: Everything Underneath The Surface

The first two anchors are above the surface. The third is what’s actually moving them.

Jono Alderson presented a framework for this at Friends of Search 2026 that’s the cleanest articulation I’ve seen. He calls it six structural capabilities of competitiveness: experience integrity, physical availability, mental availability, distinctiveness, reputation, and commercial proof.

Don’t panic. We don’t all need six new dashboards…right away.

His point is that visibility is a reflection of those signals. AI systems aggregate them across the web. When they strengthen, you become easier to find. When they don’t, no amount of prompt tracking is going to fix it.

Most of those six aren’t SEO’s job to drive. They’re product, support, brand, PR, sales, distribution. They are SEO’s job to measure, and, increasingly, to make sure our digital footprint is legible enough that machines can read all those signals accurately and in a way that’s reflective of reality. Not skewed by a technical error or lack of clear implementation. 

It’s an additional and important layer added to an old practice on the web: online reputation management. 

The supporting metrics worth watching, drawn from across his six dimensions:

• Infrastructure reliability. Not just in the “this works” way – but in the “this experience makes sense” kind of way. True UI. Throw conversion metrics in here if you want to, and they’re reliable (and closed loop). 
• Brand mentions. This can be used as a digital representative of physical availability – basically: how frequently are you in the comparative set within your industry? Are people even actually thinking of you? 
• Direct demand. Branded direct traffic, watched directionally. (direct) is now soaking up Discover, AI Mode and Overviews. You won’t know which slice is which, but if (direct) is up alongside branded search, the trend is probably real.
• Reviews and reputation trends. Volume, recency, sentiment. Stability. A brand that says different things in different places confuses both customers and machines.
• Repeat purchase and retention. Customers coming back unprompted is the cleanest commercial-proof signal you have. Hard to fake. Hard to misread.

Ideally, we do want to be measuring the whole funnel — but perfect should not be the enemy of the good with this metrics migration. Pick the two or three that match what your business is actually trying to do and watch them quarterly. 

Lily Ray got to the same place from a different angle at Affiliate Summit West earlier this year: brand visibility becomes the new KPI; influence over traffic. Jes Scholz, in the Search Engine Land piece we already cited, lands a similar diagnosis from the analytics-stack side.

Three different framings. Same conclusion. The dashboard you’re optimising for has moved underneath you, and most of the new tools are pretending it hasn’t.

I don’t know of a platform that does this cleanly across all the signals, so until we build it, we’re back to a more manual method of tracking visibility that’s the antithesis of what the AI tracking platforms are trying to direct us towards. 

Measure The Movement, Not The Position

One thing connects all of those metrics: they’re useless as absolutes.

“Our share of voice is 14%” doesn’t mean anything. 14% of what? Compared to whom? Same problem as “we rank #3.” It sounds precise, yet tells you nothing about whether anything is getting better.

What works is incremental change over time, against your goals, target and market.

“We grew excess share of voice by 10% while our closest competitors held at 2%.” 

That’s a story. That’s defensible. That’s how every other department in the business already reports — sales talks about revenue growth, not revenue. Product talks about adoption rate, not install count. Finance talks about margin improvement, not margin.

SEO has been weirdly resistant to this. We report positions and traffic totals and keyword counts as if the snapshot is the point.

It’s not. The trajectory is.

And the trajectory only means something relative to who you’re competing against.

This protects you when the absolutes look ugly. Sessions down 5% but the market down 15% — you outperformed. Branded search up 8% while your main competitor’s grew 1% — you’re winning the consideration game. TAM capture moved from 12% to 16% in a contracting market — that’s a gain in a shrinking pool, and any business leader understands what that means because they’ve seen it everywhere else.

The inverse is also true. SOV up 3% while competitors grew 12% means you’re losing ground, regardless of the positive number on your slide. Incremental framing doesn’t just make good news land better. It makes bad news actionable, because it tells you where you’re falling behind rather than just that the number went down.

This is also where Binet’s actual finding matters. It’s not share of voice that predicts growth. It’s excess share of voice — the gap between your SOV and your share of market. Track the gap. Report the gap. That’s the metric that correlates with what happens next.

The Industry’s Wrong Answer Is Already Being Sold

You’ve seen the dashboards. AI visibility scores. Brand mention frequency in Perplexity. Citation rate in ChatGPT. Confidence intervals on probabilistic outputs, sold as deterministic numbers your CMO can put in a slide.

Some of this is useful. Most of it is the same trick we played on ourselves with last-click. Take an opaque system, build a tracker that produces a number, watch the number, optimise the number, congratulate ourselves when the number goes up.

Jono called it a square peg in a different hole at Friends of Search. He’s right. Replacing rankings with AI visibility scores doesn’t solve the problem. It just buys you another year before the metric drifts and we have to do this conversation again.

Jamie Indigo went harder at Tech SEO Connect. Stop relying on synthetic AI rank tracking, they said. They called it the Lighthouse of AI — looks scientific, isn’t. Use first-party log files for real user data instead. Partner with your SREs if you have them. They’re the ones who actually know the crawl.

The problem isn’t that any of these tools are useless. It’s that most of them are quietly conflating two very different things.

Rand Fishkin and Patrick O’Donnell at SparkToro actually ran the experiment. 600 volunteers, 12 prompts, 2,961 runs across ChatGPT, Claude, and Google AI. Their findings split cleanly.

Rank position in AI? Full of baloney, in Rand’s words. The same prompt gives back the same list of brands less than 1 in 100 times. Same list in the same order, less than 1 in 1,000. Anyone selling you a “your brand ranks #4 in ChatGPT” number is selling you noise.

But brand presence frequency — how often you appear in the consideration set across many runs — held up under scrutiny. In tight categories, top brands showed up in 55-77% of responses regardless of how the prompt was phrased. Done properly — enough runs, account for weekly drift, realistic prompts, no pretending the models are equivalent — you can measure visibility with real statistical rigor.

That’s the distinction the dashboards have been blurring. Rank tracking in AI: theatre. Presence frequency at sample size: actually informative.

Most of what’s being sold is quietly the first thing dressed up as the second.

The honest version is: pick a defined query set, run it enough times to be statistically meaningful, report presence frequency over time, ignore absolute rank, and don’t make budget decisions on a single snapshot. Like a panel survey. Not a fact.

If your tooling can’t tell you that — and walk you through the methodology when you ask — it’s not the tool you need. And honestly, I’m not sure I could name one “AI tracking” tool that does that now, today, in 2026.

The Harder Conversation

Most of the trouble with new measurement isn’t the measurement. It’s that someone’s bonus is tied to the old one.

Sales has a number. Marketing has a number. Probably one of those numbers is conversions or sessions or some attributable revenue figure tied to a channel. Changing the measurement protocol mid-financial-year tells someone that the number they’ve been chasing is no longer the number they’re paid on. Doesn’t matter how right the new metric is. They’re going to fight you.

Time the change with the financial year. Or run the new metric in parallel for a quarter or two before proposing the cutover.

The other half is that finance and the board want one number. They’ve always wanted one number. The temptation is to pick a new one and present it like the old one.

Ross Hudgen’s blunt advice at Tech SEO Connect was to cut all traffic projections in half for 2026. That’s also useful framing for finance. You’re not throwing out the model. You’re acknowledging that the input has changed by half, and rebuilding around what’s still trustworthy.

That instinct to find one number is wrong but it’s not stupid. Tom Critchlow’s framing is the one I keep coming back to: you’re looking for explainability, not truth. If you can explain in a sentence what your number is and why it moves, it’s good enough to start the conversation. If you can’t, it’s not. 

That’s also how you decide what makes the dashboard.

Ian Lurie’s progressive detail framework (influenced by Edward Tufte) is the cleanest version of this I’ve found. Three layers. The top layer is for the CxO or the board — pure dashboard, no drill-down, the one or two metrics that map directly to business outcomes. The middle layer is for the VPs and directors — the same dashboard plus the supporting metrics that match their function. The bottom layer is the operating data your team works in: rankings, queries, surface-by-surface breakdowns, the messy guts of it. The level-of-detail for the people who understand while this data is still representative, they need to use it to execute the tactical work.

Don’t make one dashboard serve everyone.

And translation of the baseline metrics — explaining what’s moving, why, and whether it matters — becomes a more important skill than the dashboard itself. The number doesn’t carry the meaning. The context does.

Less Sexy. Also Less Wrong.

This is harder work. You won’t have the clean “we ranked #1 and revenue went up 20%” story. You’ll have a directional trend, a couple of proof-of-concept tests, and a stakeholder conversation that takes longer than you’d like.

You’ll also be telling the truth.

The dashboards we built in the last decade weren’t really measuring what we said they were measuring. We dressed up correlation as causation, called it attribution, and rode the numbers as long as they kept going up. AI didn’t break that. It revealed it.

So pick the anchors. Wrap the proxies around them. Be honest with finance about what you don’t know. Time the conversation with planning for the next financial year.

And when someone asks why the number isn’t doing what it used to, tell them the number wasn’t doing what we said it was doing in the first place.

---

---