SEO Share of Voice: The Number Your Competitors Cannot Hide

The slide is already on the screen when the question lands. Your competitor's name appears beside more of those results than yours does, and whoever built the slide did not need anyone's permission to publish it. Nobody in the room is wrong about what they are looking at. They are wrong about what the number underneath it is allowed to mean, and that number is about to be attached to next quarter's budget.

You have probably met the definition already. Ask one of the AI surfaces and it hands you a formula with a bracket in the middle: your organic clicks, impressions, or visibility score, divided by the total market or competitor organic visibility score, times one hundred. Three numerators in one line, and no instruction about which one to use.

I have seen two presentations of the same market, in the same month, reach opposite conclusions on identical data. That bracket is the whole argument, because the choice inside it decides who is ahead, and it gets made by whoever configured the dashboard.

What Share of Voice Measures, and What It Does Not

Share of voice is a ratio with one job: it tells you what portion of a defined conversation belongs to you. Advertising has used it for decades with a numerator made of money, media spend as a percentage of all category media expenditure in a market, channel and period. Nielsen states that definition, and states that share of voice is not an outcome metric. The search version keeps the shape and loses the units. Spend divided by spend already sits beside market share, while attention divided by attention does not, because attention is not denominated in dollars.

What the SEO version cannot tell you is whether the conversation was worth having. A rising share of a set nobody buys from measures the wrong thing accurately.

The Weighting Decision Every Page Leaves Open

A weighted index earns suspicion for a good reason, and the marketing KPIs guide sets out the test a metric has to pass before it gets a slot. This is a narrower question: the metric is already chosen, and what is left is which weight goes inside it.

Search the term and you will find the choice acknowledged, usually in one sentence, usually as house style. One guide notes that impressions are typical and clicks are possible, then moves to a how-to list. Another names weighted visibility and never states the weighting function. A third tabulates four platforms' methods and concludes that consistency beats the choice. None treats the numerator as a decision with a cost.

The objection worth answering first is the one that guide already makes: a score nobody can reconstruct on a whiteboard gets rejected. Someone always asks how their number rolled in. The objection is correct, and it sets the bar for everything below.

Weighting gets discussed as one decision. It is two: the numerator decides what counts as presence, the denominator decides whose presence counts. Against three named rivals you get one share, against every result on the page you get another.

Ahrefs' own help documentation makes the point without meaning to. Its five listed positions total 78.8% of the clicks it divides by, and the remainder belongs to results it never listed. That denominator is a decision the document does not disclose.

Impressions-based: the least work to compute, and the most flattering

Count every search in the set where you appear anywhere in the results, then divide by the same count for everyone you track. No curve, no transform, almost no computation, and the most forgiving of a thin footprint.

It rewards presence and nothing else. Rank 1 and rank 9 contribute identically, so thin coverage and one dominant page look the same. The numerator also grows whenever a new page appears on a term in the set, whether or not it earns a click. If your question is how much of this conversation you appear in, impressions answers it. If the question is whether you are winning, it does not.

Click-based: the honest one, and the curve it inherits

Clicks are the numerator I would defend. Clicks are what a business can spend, and the method separates rank 1 from rank 9. It also imports an assumption most pages do not disclose: turning a rank into clicks needs a click-through curve.

Advanced Web Ranking's July 2026 study supplies a defensible one, drawn from more than 1.75 million US SERPs per month across the top 20 positions. Position 1 earns 20.02% across all SERP types and 9.69% where an AI Overview appears, and position 2 falls from 10.36% to 3.69%. AI Overviews appeared on 43% of sampled desktop SERPs.

Now the part that should bother you. A table widely attributed to a Backlinko analysis of four million results puts position 1 at 27.6%, 38% higher than Advanced Web Ranking's figure for the same row. The identical ten-row table is also published by benchmarketing.org, sourced there to aggregated Search Console data for non-branded keywords. Two provenances, one set of numbers. A benchmarking roundup of 2026 estimates puts position 1 anywhere from about 14.6% to 39.8%.

That is a live disagreement about the first row of your multiplier, and a clicks-based share inherits all of it. Switch curves and your number moves while your rankings stay put.

Then there is the shrinking numerator. Pew Research Center tracked 900 US adults across 68,879 searches in March 2025 and found they clicked a traditional result on 8% of visits where an AI summary appeared, against 15% where none did. Fewer clicks, fewer chances for a click-based share to stay stable. The decline is not a straight line. Across 53 brands and 5.47 million tracked queries, Seer Interactive measured organic CTR on AI Overview queries bottoming at 1.31% in December 2025 and rebounding to 2.36% in February 2026. Clicks remain the right numerator, and it now partly measures whether Google answered without you. If that describes your market, AI visibility tracking is the second series to keep beside this one.

Position-weighted: what it adds, and what it distorts

The third method keeps what the first two throw away. Transform every rank into a score, sum the scores across the set, divide by the same sum for your competitors. With inverse rank, first place counts as 1.0 and fourth as 0.25, which rewards a good rank across the whole set over owning one term.

Two problems travel with it. An average position is a number, not a percentage, so turning it into a share means choosing a transform. Inverse rank, exponential decay and a CTR curve each hand you a different share from identical rankings. It is also blind to demand: rank 3 on a term with heavy demand scores the same as rank 3 on a rare one. Use it for consistency, never for reach.

A worked example where the lead changes hands

Everything here is illustrative: ten keywords at 1,000 searches a month each, two sites, one month, no client and no real domain. The click-through values are the one real input, from Advanced Web Ranking's July 2026 US desktop data.

You rank first on five of the ten keywords and are absent from the other five. Your rival ranks fourth on all ten.

On impressions, your rival appears across twice as much of the set as you do. You hold 5,000 of a 15,000 total, or 33.3%. The rival holds 10,000 of 15,000, or 66.7%. You are losing badly.

On clicks, at 20.02% for first place and 1.71% for fourth, you earn 5 x 1,000 x 20.02%, which is 1,001 clicks. The rival earns 10 x 1,000 x 1.71%, which is 171. You hold 1,001 of 1,172, or 85.4%. The rival holds 14.6%. You are winning roughly six to one.

On inverse rank, 1.0 for first and 0.25 for fourth, you score 5.0 and the rival scores 2.5. You hold 66.7% and the rival 33.3%. You are winning, by a third margin.

Same ten keywords, same month, same two sites. The lead changes hands between the first two rows, and nothing about the market changed, because nothing about the market was ever in the calculation. Only the numerator moved.

Now run the other published curve over the same clicks method, 27.6% at first place and 8.4% at fourth. You earn 1,380 clicks, the rival 840, and your share falls from 85.4% to 62.2%. Your ranks did not move. A different curve, on identical data, took 23 points off your result.

Which brings us back to the whiteboard objection. Every figure above comes from four inputs: the keyword count, searches per keyword, each site's rank, and one named curve. All four fit on a whiteboard, and a colleague can reproduce 85.4% in a minute with a calculator. A weighted index fails when the weight is hidden, and survives when the weight is four numbers and a published curve anyone can check. Then the argument stops being about whether your number is real and becomes about which question you want answered.

Share of Voice versus Share of Market

Share of market is your revenue as a percentage of category revenue. Share of voice, in its advertising form, is your media spend as a percentage of category media spend. Subtract the first from the second and you have excess share of voice, the quantity the Binet and Field work is built on. Nielsen reports that study as 171 campaigns across top categories between 1980 and 2010, with market share gaining 0.5% for every 10% of excess share of voice.

The rule is computed from spend, so it applies to spend. Here is the wall an SEO share of voice hits: you cannot subtract a clicks-based share of voice from a revenue-based share of market, because the units do not match. Do it anyway and you make a budget decision with a number nobody can name.

The honest bridge is one assumption, stated out loud. An attention share becomes a revenue share when you multiply it by the ratio of your revenue per unit of attention to the category's. If your site monetizes a visit at the same rate as the average competitor, the two converge. Every remaining gap is then a monetization difference worth reading: it tells you whether you convert the attention you own or merely appear in a conversation that does not pay.

Keep the two apart in your head, and name the conversion whenever you put them on one slide. Share of voice is how much of the conversation you own. Share of market is how much of the money you own. The distance between them is a strategy question wearing arithmetic as a disguise.

How to Build the Series Without a Vendor

Which platform to run this on is a separate and better-worn question, and the marketing analytics tools guide maps it layer by layer. Whichever platform produces the number, the number carries the same two choices.

The minimum keyword set

The set is a decision, not a default you inherit from a tracker. Every term should be one a buyer would type before they know your name, because a term only your existing customers search measures brand recall, not the market. And every term should have at least two competitors in the top ten, because a term you win by default inflates your numerator for free.

Then freeze the membership. Adding ten terms you dominate moves your share further than switching methods will, and on a chart it looks like growth. Keep the list beside the method, and date any change. A number computed over a moving set is not a series.

The one chart that survives a re-org

Plot one line per site over the longest window you have, from one method, over one frozen set, with the method in the subtitle and the curve credited underneath. That chart outlives the person who built it: anyone who opens it sees which numerator drew the line, and anyone who disagrees can recompute a point.

Two rules keep it alive. Never stack two methods on one chart, because the break gets read as a market event. Never rescale the axis when a competitor enters, because the rescale becomes the story.

How often to send it and in what shape is a solved problem elsewhere in this cluster, and the SEO client reporting guide covers it. This article stops at the series itself: the set you measure, and the one chart worth keeping after the team that built it has moved on.

Conclusion

The number is not the hard part. Three methods, one keyword set and a calculator produce three answers in ten minutes, and only one can sit beside your market share afterward. What it cannot do is decide anything by itself: not whether those terms were worth owning, and not whether a competitor appearing everywhere is a threat or a rounding error.

What it can do is narrow a question. Fix the set, name the numerator, credit the curve, and you have something that survives the meeting where someone asks how it rolled in.