การวัดผล
The new share of voice, for the AI era
For decades, public relations ran on one scoreboard: share of voice. It measured your slice of the conversation, the proportion of coverage in your category that carried your name against everyone else's. It was countable, it was comparable, and for a long time it tracked something real. If people were going to hear about you, they heard through the media, so counting the media was a fair stand-in for mindshare.
That stand-in is coming apart. Your buyer no longer starts with a search box or a trade magazine. They open an assistant and ask a direct question. Who are the best options for a company like mine? Which of these two should I pick? The assistant reads widely and answers briefly, naming a few names. That short answer is where the decision now begins, and often where it ends.
This book is about the metric that replaces share of voice in that world. Call it share of answer: of the answers assistants give to the questions your buyers ask, how often are you named, how well, and against whom. The chapters ahead define it, give you a free way to measure it every month, explain what moves it, treat it as the competitive game it is, and show you how to operate on it month after month. Throughout, it stays honest about what these answers can and cannot tell you. The aim is a metric you can run, not a phrase you can repeat.
Chapter 1
Share of voice began as a media-buying idea and became the default scorecard for earned media. The question it answered was simple. Of all the coverage in your category this month, what fraction was about you? Count the articles, the broadcast segments, later the social posts, weigh them if you like, then divide your pile by the total. A bigger slice meant more people were hearing your name than your rivals' names. More mindshare, more consideration, more sales. That was the chain of logic, and for decades it mostly held.
It held because the buyer's path ran straight through the media you were counting. A trade journalist wrote about a problem and named three vendors. A buyer read the piece, recognized the names, and built a shortlist. The coverage and the decision were links in one chain. Counting the coverage was a fair way to estimate your place in the decision, because the decision was assembled out of the coverage.
That chain has been cut in the middle. Your buyer still has the problem. They no longer begin by reading around it. They ask an assistant. Who are the best payroll tools for a small company in Singapore? Is this vendor or that one better for a team like mine? What are the alternatives to the incumbent everyone mentions? The assistant does not hand over a reading list. It returns an answer, and the answer names a few options.
This shift reaches past one assistant or one kind of buyer. A procurement lead, a founder, a marketing manager, each increasingly treats an assistant as the first stop, the way a search engine was the first stop a decade ago. The difference that matters for measurement is the shape of what comes back. A search engine returned a page of links and left the choosing to the buyer. An assistant does some of the choosing first, then hands over a shortlist. Measuring the links you ranked for told you something. Measuring the shortlist tells you far more.
The old metric also flattered breadth over trust. A hundred small mentions outscored a single authoritative one, because the count never asked who was listening or whether anyone believed the source. The answer inverts that. It rewards being the name a trusted source vouches for and ignores the pile that no one credible stood behind. A company tuned for raw mention volume can be perfectly optimized for a scoreboard that no longer decides anything, winning a race that buyers stopped watching.
Here is the break. You can win share of voice and lose the answer. You can have more articles than a competitor this quarter, a bigger slice of the clippings, a healthier monitoring report, and still go unnamed when a buyer asks the assistant the one question that precedes the purchase. The slice you are counting and the moment that decides the sale have come apart. The proxy no longer points at the thing it stood in for.
Assistants compress. They read a great deal and say very little. Volume is one signal among many, and a weak one. A single source the model trusts can outweigh fifty thin ones it does not. Fifty press releases nobody credible picked up can leave you absent from an answer that names a rival on the strength of one well-placed article. The arithmetic of share of voice scores those fifty as a win. The answer treats them as noise.
There is a second problem. Much of the new decision happens where no clipping service looks. When an assistant names three vendors to a buyer, nothing is published. There is no article, no clip, no entry in your monitoring tool. The most important impression of the month, the one that shaped a live buyer's shortlist, leaves no trace in the system you built to track impressions. Your dashboard can look quiet on the exact day you are losing.
There is a timing problem on top of that. Share of voice reported last month, after the clips were gathered and tallied. The answer is live. It changes the day a new source is published or a model is refreshed, and a buyer gets the current version, never the one your report described. You were reading a photograph of the past while the thing that matters updates on its own clock. By the time the monthly clipping report lands, the answer a buyer saw this morning has already moved.
Put the two problems together and the old report becomes unreliable in both directions. It can show a strong month while you are losing answers, and it can sit flat while you are quietly winning them. A measure that can move opposite to reality is worse than no measure, because it hands you false confidence. That is the real cost of clinging to share of voice now. It does not only miss the new moment. It can point you the wrong way.
Consider how this feels in practice. A founder earns a strong writeup in a respected outlet and a wave of smaller pickups. The share of voice report looks excellent that month. The same founder opens an assistant, asks the obvious buying question in their own market, and watches it name two competitors and leave them out. Both facts are true at once. The coverage was real, and the absence from the answer was real. Only one of them maps to the buyer standing at the decision.
None of this means PR is finished. Coverage matters more now, because coverage is one of the things the assistant reads before it answers. The work of earning trust in credible sources is more valuable than ever. What is finishing is the metric. Counting mentions tells you less every quarter about whether your name is in the room when your buyer decides. The activity is healthy. The old scoreboard is quietly dying, and keeping faith with it is starting to cost real decisions.
Takeaway: Share of voice still counts what gets published. It no longer counts where your buyer decides.
Chapter 2
If share of voice is dying, something has to replace it, and the replacement has to describe the new moment precisely. That moment is an assistant answering a buying question with a short list of names. The metric for it is share of answer: of the answers assistants give to the questions your buyers ask, the share in which you appear, weighted by how well you appear and measured against the rivals who appear beside you.
A retail shelf is a useful picture. On a shelf, being stocked beats sitting in the stockroom. Eye level beats the bottom row. A clean facing with a clear label beats a crushed box turned sideways. Being the only one of your kind in view beats being one of nine. The AI answer is a shelf with very few slots and a buyer standing right in front of it, asking the staff what to buy. Share of answer measures your place on that shelf.
The picture has one more twist. The buyer trusts the staff. When a person runs a search and sees ten links, they know they are choosing for themselves, and they weigh the source as they go. When an assistant names three options, the recommendation carries borrowed authority, because the buyer treats the answer as considered advice rather than a list to sift. That makes a slot in the answer worth more than a slot on a results page ever was. Being named is close to being vouched for, and being left out reads as not making the cut.
The metric breaks into four parts. Keeping them separate is what turns it into something you can track instead of a feeling you carry around.
Two companies can post identical appearance rates and see very different outcomes. One is named first and called the clear leader. The other is named last and tagged as a budget option with a caveat. The binary says they tie. Position and framing say they do not. This is why the metric has to carry more than a yes or no, and why recording how you are described matters as much as recording that you showed up.
Put the four together and share of answer stops being a vague yes or no. It is an appearance rate across the questions your buyers ask, adjusted for where and how you appear, read against the competitors who share the shelf with you.
Share of voice measured mentions anywhere and at any time, most of them far from a purchase. A profile in a feature article is a mention. So is a passing reference in a roundup nobody acted on. Share of answer measures a narrow, valuable moment: a buyer asking a buying question and receiving a shortlist. That is closer to the sale than almost anything the old metric counted. You are reading the shelf at the instant the buyer reaches for something.
There is also a difference in intent. A reader who stumbles on your name in an article may or may not be in the market. A buyer who asks an assistant which option to choose has raised their hand. They are close to a decision and looking for a recommendation. Being named in that reply reaches someone who is actively deciding, which is the audience PR always wanted and could rarely confirm it was reaching.
One more contrast with the old metric matters. Share of voice was additive. Every mention you earned raised your number and took nothing from anyone, which is why it rarely felt like a fight. Share of answer is positional. The room is fixed, the names are few, and your gain is usually someone's loss. An additive metric invites more activity for its own sake. A positional metric forces choices about which answers are worth winning and which to let go. That difference changes how you work, and most of this book follows from it.
The word share is doing real work. The answer has limited room by design. An assistant that named fifteen options would be useless, so it names few. Every name that makes the list keeps another name off it. The metric is relative by construction. Your absence is literally a competitor's presence in the same sentence. That is why it behaves like a game, which is the subject of a later chapter.
The answer is not carved in stone. Ask the same question twice and the wording shifts. Sometimes a name drops in or out. This is a real property of the thing you are measuring, and any serious method has to account for it. Share of answer is best understood as a distribution you read over repetition, never a single reading you take once and trust. The next chapter turns that caution into a disciplined habit instead of an excuse to look away.
Takeaway: Share of answer is appearance, position, framing, and competitive share, read at the moment your buyer asks.
Chapter 3
You can measure share of answer yourself, for free, starting this month, with a spreadsheet and an hour. No product is required to begin, and beginning by hand is the right move, because the discipline is what makes the number trustworthy. A tool can speed this up later. It cannot do the first month's thinking for you.
Start with the questions a real buyer would type, in their own words, at the point of choosing. These are not keywords. They are full questions. Best option for a company like mine in my city. Alternatives to the incumbent everyone names. Whether vendor A or vendor B is better for my use case. Which provider is cheapest for a small team. Write fifteen to thirty of them. Phrase them the way a buyer actually talks, because the assistant answers the real phrasing, never the tidy marketing version.
Spread the questions across the ways buyers actually search. Some are broad and early, asking who the main options even are. Some are narrow and late, comparing two named finalists. Some name a problem rather than a category, the way a buyer describes a symptom before they know the cure. A set that is all broad questions will either flatter you or bury you and teach you little either way. Aim for a mix that mirrors the real path from first look to final choice, because the answers differ sharply along it, and so does what it takes to win them.
In Asia, write them in more than one language. A buyer in Jakarta may ask in Bahasa Indonesia. A buyer in Bangkok may ask in Thai. The same question in English and in the local language can return different names, drawn from different sources. If your buyers ask in two languages, your question set needs both, or you are measuring half your market.
Ask every question in each of the major assistants your buyers use: ChatGPT, Claude, Perplexity, Gemini. They read different things and answer differently, so a name that appears in one can be missing from another. Use a clean session, signed out or in a fresh state, so your own history does not teach the assistant to flatter you. You want the answer a stranger would get, not the answer tuned to you.
Keep one row per question per assistant in a simple sheet. Columns that work: the question, the assistant, the date, whether you were named, your position, a short phrase for how you were framed, and the full list of competitors named. That last column is the one most people skip and the one that pays off most, because it hands you your real competitive set and your target list in the same stroke.
Because answers wobble, do not trust a single pull. Ask each question two or three times and record each result. Your appearance rate is the fraction of all those readings in which you were named, not the outcome of one lucky or unlucky attempt. Repetition is what converts a noisy single answer into a number you can stand behind.
Turn the sheet into a few stable numbers. A workable scorecard: appearance rate as the headline, the share of readings that named you. Then an average position. Then a framing tally, how often you were cast as leader, as cheap, as risky, as safe. Then a competitor leaderboard, which rivals appeared most across your questions. The exact formula matters far less than keeping it identical month to month. A rough score you never change beats a precise score you redesign every time.
A worked version makes this concrete. Say you track twenty questions across four assistants and ask each one three times. That is two hundred and forty readings for the month. If your name appears in ninety of them, your appearance rate is thirty-eight percent, and that single figure is your headline. Of those ninety appearances, count how many put you in the top group and how many left you trailing, and you have your position read. Tally the adjectives attached to your name and you have your framing. List every rival named, sort by how often each appears, and you have your leaderboard. None of it needs math beyond counting, which is the point. The rigor comes from doing it the same way every month, not from the sophistication of the formula.
Fix everything you can. Same questions, same assistants, same method, the same week of each month. Change the method and you lose the comparison, which is the entire value. When you want to add new questions, track them as a separate set so your core trend stays clean. Note the date and any model version you can see, because answers move when models update, and you will want to tell your own progress apart from the model's churn.
This does not need a specialist. A founder or a single marketer can run the whole cycle in an hour or two a month once the question set exists. The first month is the slow one, because you are writing the questions and building the sheet. Every month after is mostly repetition, which is exactly what makes it trustworthy.
Be honest about what this is. It is a sample, not a census. You cannot observe every buyer's phrasing or every answer they receive. You are reading a representative panel of questions, repeated, over time. No single percentage is the truth of your standing. The trend across months and the comparison against rivals are where the signal lives. Held that way, a handmade scorecard tells you more about where you stand than any clipping report ever did.
Takeaway: Same questions, same assistants, repeated monthly and recorded the same way. The trend is the truth, not any single answer.
Chapter 4
Once you can see your share of answer, the next question is what raises it. The inputs are knowable, and they can be ranked by leverage. Before the ranking, one honest statement: all of these take time, none is a switch you flip, and anyone promising to move your share of answer this week is selling something. With that said, here is what matters, strongest first.
This is the highest-leverage input, and it is the oldest work in PR aimed at a new reader. Assistants weight their sources. Being named and described inside a publication the model treats as credible does more than a pile of mentions in places it does not trust. One substantive article in a respected national or trade outlet, where you are framed as a real option, can change how an assistant names you. The model is now part of the audience for every placement. A placement that moves a buyer and a placement that moves an answer are increasingly the same placement.
The model builds its picture of you from everything it reads. When your category, your positioning, and your name are described the same way across your own site, your profiles, and the coverage about you, the model forms a confident picture and repeats it. When your sources disagree about what you even do, the model hedges, softens, or drops you, because it will not confidently recommend something it cannot cleanly describe. Consistency is close to free, and it compounds quietly. For a company whose story is currently muddled, it is often the fastest gain on this list.
Models draw on a wide field: reference sites, reputable directories, databases, the broader open web. If your category has sources the models lean on, you need an accurate and current presence in them. This is about being legibly present in the places the answer is assembled from, with facts that match your consistent story. It rewards being easy to find and easy to describe correctly, and it punishes stale or contradictory entries you have forgotten about.
For Asia this is leverage, not a nicety. A buyer asking in Thai, Vietnamese, or Bahasa receives an answer shaped largely by sources in that language. Coverage and presence in the buyer's own language move the answer for that buyer in a way English coverage does not reach. Few competitors do this well, which means the local-language shelf often sits open while everyone crowds the English one. For a company serious about a specific Asian market, this can be the highest-return work available, even though it sits fourth in the general ranking.
That ranking is a general order, and your own order depends on your biggest gap. If your sources contradict each other about what you do, consistency jumps to first, because no amount of coverage helps while the model cannot describe you cleanly. If you are credible in English and invisible in a target language, in-language work moves to the top for that market. Use your measurement to find the binding constraint, the one weakness holding the whole number down, then spend there. Leverage is highest wherever your current weakness is most severe, so read your own scorecard before you copy anyone's playbook.
Volume for its own sake does little. Keyword stuffing does little. Thin press releases that no credible source picks up do little. Most of all, your own claims about yourself do less than you want them to. Models discount what a brand says about itself and lean on what others say about it. This is the hard truth of the metric: you cannot simply write your own answer. You earn it through the people and sources the model already trusts.
There is a delay between input and result. Models refresh their view of the world on their own timetable. A strong piece of coverage might not surface in answers for weeks or longer. Treat your inputs as planting and your monthly measurement as the harvest, read seasons apart rather than days apart. Judge the work on a quarter, not on next Tuesday's pull. The teams that win this are the ones who keep planting while the impatient ones give up and declare it broken.
Measure the inputs, not only the outcome. Track whether the consistent story is actually live everywhere, whether the trusted pieces ran, whether the in-language coverage exists at all. When your share of answer moves, you want to know which input moved it. When it does not move, you want to know whether you did the work or only planned it. An input you cannot confirm you shipped is no evidence of anything. This is how you keep the loop honest and stop yourself blaming the model for work that never left the building.
Then accept the part nobody wants to hear. You cannot buy this number directly. There is no placement, no page, no payment that writes your name into the answer on demand. You assemble the conditions under which the model names you, and then you wait for it to. That is slower than a campaign and sturdier than one, because a position earned this way is harder for a rival to knock out than a burst of coverage ever was. The slowness is the moat. Anything you could buy in a week, a competitor could buy in a week too.
The through-line is simple. The things that move share of answer are the things that have always built reputation: credible third-party coverage, one clear and consistent story, accurate presence in trusted places, delivered in your buyers' languages. The difference now is that a second reader, the model, stands behind every one of them, and it is the reader deciding which names reach your buyer.
Takeaway: Credible coverage, a consistent story, presence where models read, in your buyers' languages. In that order, and over months.
Chapter 5
Share of answer behaves like a game because it is close to zero-sum for each question. The answer holds a handful of slots. When a rival is named and you are not, you did not merely underperform in the abstract. You lost a specific slot to a specific competitor, on a specific question, in a specific market and language. Reading it as a contest, with named opponents and a scoreboard, is the right posture.
Your monthly measurement already records who appears beside you and who appears when you are absent. That list is your true competitive set, the one the model recognizes, and it may not match the one in your pitch deck. The company you treat as your rival may rarely appear in the answers your buyers receive. Another company you barely track may take slot after slot. Believe the answers. The names that keep showing up are the ones to study: what is said about them, and which sources name them. That tells you where each slot was actually won.
The most valuable output of measurement is the list of questions where a competitor is named every time and you never are. Do not read that list as failures to mourn. Read it as a target list. Each lost question is concrete. It has a specific answer, built from specific sources, saying specific things. You can see what the model is reading and repeating, which means you can decide what coverage would earn you a place in it. A vague goal of more visibility becomes a short, ranked list of exact answers to go win.
Sort the losses before you chase them. A question you lose where the buyer is ready to decide is an emergency. A question you lose that no buyer acts on is a footnote. Group your lost questions by how close they sit to a purchase and by how winnable they look, given who holds the slot now and why they hold it. The top of that sorted list is your PR brief for the next quarter, written for you by your own buyers and the assistants they trust. It is a sharper brief than most teams ever write for themselves.
You will not win every question, and trying to is a reliable way to win none of them. Choose the questions closest to a real buying decision and closest to your genuine strength. Being named in the answer to the highest-intent question in your core market is worth more than appearing in a broad definitional query no buyer acts on. Concentrate your coverage and effort where being named changes a sale. A few won questions that matter beat a scattered presence across questions that do not.
Picking fights also means knowing when a slot is not worth contesting. Some questions are dominated by a giant for reasons you will not overturn this year. Some are so generic that being named changes nothing. Spend your energy where the gap between you and the winner is closeable and where closing it touches revenue. The list of questions you ignore on purpose is as much a part of the strategy as the list you chase.
This is where the game gets specific to the region. Share of answer is not one scoreboard in Asia. It is many. Best option in Singapore English, in Hong Kong, in Bahasa Indonesia, in Vietnamese, in Thai: these can return different names, drawn from different sources. A competitor can own the English answer in a market and be completely absent from the same market's local-language answer. That is a warning and an opening at once. A larger regional incumbent may dominate the English shelf while the local-language shelf sits empty. Focused in-language work can win a market on that shelf before the bigger rival even registers it as a separate board.
Timing favors the first serious mover in a language. The company that earns credible local-language coverage early becomes the name the model has on hand when buyers in that market start asking. Later arrivals have to displace an answer that already reads as settled, which is harder than filling one that was empty. In markets where no one has done the work, the opening is real and the opening is temporary. It is a window, and windows close as competitors notice the board exists. The race in Asia is won market by market, and usually by whoever treated each market as its own race first.
Slots are not kept forever. A rival's new coverage, a model update, a shift in how you are framed, any of these can drop a name that was winning last month. The questions where you appear today are a position to defend, not a trophy to shelve. This is the real reason measurement is monthly. You are watching for losses as much as gains, and a slot slipping is an early signal worth catching before it hardens into a trend.
One limit deserves stating plainly. When you study why a rival wins a slot, you are inferring from outputs. You can see that they are named and form a strong guess from the sources that name them. You cannot open the model and read its ranking or its reasons. Treat your competitive map as a well-evidenced hypothesis you update every month, never as certainty. It sharpens the longer you track it, and it beats guessing by a wide margin, as long as you remember which one it is.
Takeaway: Someone is named and someone is not. Learn the questions you lose, pick the ones that matter, and play each market and language as its own board.
Chapter 6
Share of answer earns its keep only when it becomes a monthly operating rhythm wired into your PR. As a number you admire now and then, it is a vanity score with extra steps. As a loop that changes what you pitch next month, it is the most direct read you have on whether your reputation is reaching buyers at the point of decision.
Once a month, run the same questions, record the answers, and update the scoreboard. Then read three things and only three. Your appearance rate and which way it moved. The questions you newly won or newly lost. The competitors gaining ground on you. Then take one action in response: point that month's PR and content at a chosen set of questions you are losing. That is the whole discipline. Measure, pick a target, earn coverage against it, measure again. A measurement that leads to no action is where this slides back into a vanity score.
Keep the monthly review short enough that it actually happens. Thirty minutes with the scoreboard open is plenty: confirm the appearance rate and its direction, star the three most valuable questions you are losing, name the rival taking them, and write the one pitch or piece that goes after them this month. A review that swells into a research project gets skipped the first busy month, and a loop skipped once tends to stay skipped. Small and repeated beats thorough and abandoned, every time.
Every pitch, every placement, every founder interview is now also an input to an answer. Ask of each one which question it helps you win, in which market, in which language. A placement chosen that way does two jobs at once: the traditional job of reaching a human reader, and the new job of lifting a specific answer for the next buyer who asks. The scoreboard tells you which placements to chase. The PR work moves the scoreboard. Neither half is worth much alone.
Give the loop an owner and a place on the calendar. It can be the founder for a while, then a marketer, then a tool, in that order as you grow. What cannot happen is for it to be nobody's job. A metric with no owner is a metric that gets measured twice and then forgotten the quarter it starts to matter.
Guard against the metric becoming a vanity score in a new costume. The test is plain: can you name the sale that being in an answer helped win? Often you cannot draw a clean line, because a buyer who saw your name in an answer may never tell you so. What you can do is keep the metric pointed at the questions closest to revenue and treat movement there as the number that counts. A rising appearance rate on questions no buyer asks is a prettier vanity score, nothing more. Chase the answers that sit in front of a decision, and let the rest stay unmeasured.
Anyone who hides the limits of this metric is overselling it, so here they are. Answers vary between identical questions, which is why you read rates over repetition and never trust a single pull. Phrasing changes results, so your question set is a sample of how buyers ask and never the whole of it. Models update, so a jump or a drop in your score can be the model shifting rather than you. And you are always reading outputs, never the model's reasons, so your explanations stay hypotheses. None of this makes the metric useless. It makes it a trend to track with discipline, the way you would treat any market measure, held honestly and never sold as a single clean truth.
Doing this by hand works, and you should start there. A tool earns its place when the manual version gets heavy, which it will: more questions, more languages, more assistants, repeated every month, with competitors to track across all of them. That is the work Cites X is built for. Cites X reads your share of answer across the assistants and against the competitors who share your answers, tracking appearance, position, and framing over time, and surfacing the exact questions you are losing. Because Cites X also does the brand and PR work that are the inputs, it connects each losing question to the coverage that would move it. The measurement and the lever sit in one place: see the question you lose, see who holds the slot, aim real coverage at it, and watch the next month's reading. It keeps the honest limits in view rather than selling you a single number as the truth.
Do not wait for a tool or a budget to begin. This week, write down fifteen questions a real buyer would ask an assistant before choosing a company like yours. Open ChatGPT, Claude, Perplexity, and Gemini. Ask all fifteen in each. For every answer, note whether you were named, where you sat, how you were described, and who else was listed. That single page is your first share of answer reading. It will tell you more about where you truly stand today than a month of counting mentions ever could. Everything in this book builds out from that one page, and the companies that win the AI era are the ones who make that page every month and then act on what it shows.
Takeaway: Measure monthly, aim your PR at the questions you lose, stay honest about the limits, and start this week with fifteen questions and four assistants.
Cites X runs the whole loop: build the brand, win the coverage, and become the answer AI gives.
ขอสิทธิ์เข้าใช้งาน → More Field Guides →