Online reputation metrics: what to measure and how
Online reputation metrics that hold up: page-one exposure, rating math, branded search, response time and AI accuracy, with formulas and a scorecard.
The online reputation metrics worth tracking are few: how much of page one for your name is accurate and yours, how exposed you are to the negative results that remain, what your ratings are and how fast they move, how many people search for you by name and click, how quickly you reply, and how often AI assistants state a given fact correctly. Each one is a number you can compute from a dated observation, repeat every month and compare with the last reading.
Most dashboards fail the other way. They show one blended "score", a mention count and a sentiment pie, and none tells you what to do next. A useful metric answers a question you could act on: is the problem getting smaller, is the good material taking space, are people still finding and trusting you?
This guide builds a small scorecard: the formulas, a worked example with invented numbers you can check, the rating math that most review advice skips, and the traps that flatter a bad situation. It does not teach removal or suppression, and it leaves the first-time audit and the AI checking routine to their own guides. Platform rules change, so where a point rests on a help page, the page is linked.
- Track six families of metrics: page-one exposure, branded search, ratings, response, open issues and AI claim accuracy. Skip blended scores and raw mention counts.
- Freeze the definitions on day one: the queries, the device, the location, the labels and the formula. A metric whose definition drifts cannot show a trend.
- Weight page one by position. Two negative results can stay at two for six months while your real exposure to them halves.
- Small numbers lie. One bad review on ten reviews moves a rating almost five times as far as the same review on fifty.
- Read the scorecard monthly, judge work over quarters, and write down what you would do if a number crossed a line before it does.
Online reputation metrics, and the question each one answers
A reputation metric is useful only when it answers a question that leads to an action. Everything else is decoration, however neat the chart. The table below lists the metrics this guide keeps, with the question each answers, where the data comes from, how often to read it and the trap that most often spoils it.
Related readReputation management report: what to expect| Metric | Question it answers | Source and cadence | Main trap |
|---|---|---|---|
| Page-one ownership share | How much of the first screen is accurate material you wrote or shaped? | Manual logged search, monthly | Counting slots, not positions |
| Negative exposure index | How visible are the results that hurt? | Same log, monthly | Using a different query set each time |
| Branded search clicks and CTR | Do people look for you by name, and do they click your pages? | Search Console, monthly | Reading it as a verdict on others' pages |
| Rating, volume, velocity | What do customers say, and is it moving? | Each review platform, monthly | Trusting an average on a tiny sample |
| Reply rate and first-reply time | Do you answer, and how fast? | Your own log, monthly | Fast but careless replies |
| Open issue count and age | Are known problems being closed? | Issue register, monthly | Closing the issue, not the cause |
| AI claim accuracy | How often does an assistant state a fact about you correctly? | Fixed questions, repeated runs | One run treated as the answer |
Missing on purpose: follower counts, impressions on your own posts and an all-in "reputation score". None tells you what a stranger finds after typing your name into a search box or asking an assistant. The sections below take each family in turn, starting with the discipline that makes them comparable.
Fix the baseline and the definitions before you measure anything
A metric is a comparison, so it is only as good as the first reading. Most failures are definition drift: the search in March used six queries on a phone, the one in June used three on a laptop, and a "trend" appeared that no one caused. The setup takes about an hour and happens once.
- Choose five to eight queriesUse your exact name in quotes, the name plus your city, the name plus your role or company, any older spelling, and for a business the brand name plus "reviews".
- Fix the conditionsPick one device, a private browser window with no account signed in, and one location, and record them in the log header. Aim for consistency, not a universal truth.
- Define the labelsEvery result on page one gets one label: owned (a page you control), earned positive (accurate third-party material that helps), neutral (accurate, neither helps nor harms), negative (accurate or disputed material that harms), or unrelated (someone else with your name). Write the test for each label in a sentence.
- Record each resultFor every query log the date, position, URL, label and a screenshot reference: ten rows per query.
- Set the ratings snapshotFor each review platform record the average, the number of reviews, the count of each star level and the date of the most recent review.
- Open the issue registerList each known problem with a name, the date found, an owner and a status. Close items with a date; never delete them.
A critical but fair news article counts as negative for exposure even though it is accurate and lawful: the metric measures what hurts, not what is false. Keep a short list of borderline rulings.
A first-time audit scores one moment across many areas; this baseline keeps only what you will repeat. If you have done an audit, reuse its queries and raw observations.
Page-one ownership and negative exposure: weighting by position
The most useful search metric is the share of page one that is accurate material you own or shape, next to a measure of how exposed you are to what hurts. Both come from the same log, and both need position weights, because a result at position two is seen far more often than one at position nine.
Counting slots hides this: a negative article moving from position three to six leaves the count unchanged, and a weighted index reports the gain.
Related readReputation Management RFP and Vendor ScorecardFix a weight table on day one. These weights are a convention for this guide, not a published standard, and any reasonable scale works if you never change it: positions one to three weigh 3 points each, four to six weigh 2, seven to ten weigh 1. A full page totals 19 points, and each category share is its points divided by 19.
Worked example, invented numbers: a consultant tracks one query, her name in quotes, over six months. These are not client results and imply no outcome.
| Reading | Owned | Earned positive | Neutral | Negative |
|---|---|---|---|---|
| Month 0 | 9 of 19 (47%) | 2 of 19 (11%) | 4 of 19 (21%) | 4 of 19 (21%) |
| Month 3 | 9 of 19 (47%) | 3 of 19 (16%) | 4 of 19 (21%) | 3 of 19 (16%) |
| Month 6 | 9 of 19 (47%) | 5 of 19 (26%) | 3 of 19 (16%) | 2 of 19 (11%) |
Every month has exactly two negative results, a news article and a forum thread, so a slot count says nothing changed. The weighted view shows negative exposure falling from 4 points of 19 to 2, because the article drifted from position three to six and then eight while the forum thread moved from seven to ten. The owned share did not move at all; the gain came from accurate earned material, such as interviews, taking higher positions.
Computed from the example log above with the 3, 2, 1 position weights. Not a client result and not a forecast.
Three refinements keep this honest. Average the shares over all five to eight queries. Treat a one-point change as noise, since two neighbors swapping places moves the index that much. And read the unrelated label separately: a namesake in positions two and three holds your ownership share down, which is a different problem from a negative result. Image blocks, video blocks and knowledge panels have no stable position weight, so log them as a yes or no in a notes column.
Branded search: what Search Console can and cannot tell you
Search Console shows whether people search for your name and whether your own page earns the click. It cannot show where other people's pages rank, so it complements the manual log.
Related readHow long does reputation management take?Google's Search Console help defines the terms. An impression means a user has seen, or potentially seen, a link to your site. Clicks generally count when a user goes to a page outside Google's surfaces. Click-through rate is clicks divided by impressions. Average position is the topmost position held by a link to your site, averaged across queries, as set out on Google's Performance report help page. So the position figure is the best position of your own site, not what a searcher saw overall.
Google also announced a branded queries filter in a November 2025 Search Central post. Its text could not be read for this guide, so check the current documentation for how it classifies queries and who can use it. Filtering the query list by your name works without it.
Read three things each month. Branded impressions show whether more people search for you, but a rise can follow good news or bad, so compare its date with your issue register. A falling CTR with steady impressions usually means something else is taking the click, so check the page one log first. Low clicks to your best page suggest a title and description that do not match what searchers expect.
An invented example: month one has 1,200 branded impressions and 300 clicks, a CTR of 25 percent. Month two has 1,500 impressions and 330 clicks, a CTR of 22 percent. Interest rose by 25 percent while the share who clicked fell, so check the log for what appeared above or beside your page. The report cannot say; the log can. Search Console also needs a site you own and have verified. With no site, you have no branded data, which is itself a finding.
Related readReputation management brief: a template for agenciesReview metrics: average, volume, velocity and the arithmetic of moving a rating
Four numbers describe a review profile: the average, the number of reviews, the spread of stars and the rate of new reviews. The average is the least informative alone, because its meaning depends on how many reviews stand behind it. Platforms do not all compute it the same way. Google's help page on review scores says the score is the average of all ratings published on Google for that place or business, and that a new review may take up to two weeks to be reflected in the updated score. The page as read for this guide gives no rounding rule and no recency weighting, so do not assume either. Trustpilot's trust page says each score is based on the number of reviews, how old they are and a weighting system, and that the weighting includes seven reviews at 3.5 stars so that businesses with few reviews start with a balanced score. The full formula is in a separate help article that could not be read for this guide.
Sources: Google Business Profile help, "Review scores"; Trustpilot, trust page; Google Search Console help, "Performance report". Each is linked in the text.
A plain average treats ten reviews and a thousand as the same kind of evidence; the Trustpilot approach is a prior that pulls a young profile toward the middle. As a simplified illustration, ignoring review age, three five-star reviews would sit at (15 + 24.5) divided by 10, which is 3.95, and 50 reviews averaging 4.2 at (210 + 24.5) divided by 57, about 4.11. The real formula was not on the page read, so this shows the effect, not any actual score.
Related readHow to document online defamation evidenceHow many new reviews it takes to move an average
Run this calculation before you set a target. It uses the plain average, as Google's page describes.
An invented business has 50 reviews averaging 4.2, which is a rating sum of 210. Every new five-star review adds 5 to the sum and 1 to the count. To reach 4.3 you need 8 new five-star reviews: (210 + 40) divided by 58 is 4.31, while 7 reviews give 245 divided by 57, which is 4.298 and falls just short. To reach 4.4 you need 17: (210 + 85) divided by 67 is 4.403, whereas 16 gives 4.394. To reach 4.5 you need 30: (210 + 150) divided by 80 is exactly 4.5.
Computed from a rating sum of 210 over 50 reviews. The needed count rises faster than the target because each step is harder than the last. Rounding is not modelled.
The reverse matters as much. One new one-star review on that 50-review profile gives 211 divided by 51, or 4.137, a drop of about 0.06. The same review on ten reviews averaging 4.2 (a sum of 42) gives 43 divided by 11, or 3.91, a drop of about 0.29, nearly five times larger. A young profile should watch the count and the star distribution, not only the average, and one angry review at launch fades as volume grows.
Velocity, and why a sudden spike is not a goal
Review velocity is new reviews per month; a steady trickle beats bursts. Google's contribution policy says that incentivized or biased reviews are not allowed and will be removed, and it treats content showing unusual volumes or patterns of review contributions as a violation. Merchants may ask customers to share genuine experiences but may not offer incentives or discourage negative ones, according to the policy page. Aim for a flat monthly line, never a spike from a campaign, a discount or a batch of staff-requested reviews.
Related readOnline reputation checklist for individualsAlso record the star distribution: a 4.2 of mostly fives and a few ones is a different business from one of mostly fours.
Response metrics: reply rate, first-reply time and open issue age
Response metrics measure what you control, so they are the most reliable numbers on the scorecard.
Reply rate is replies divided by reviews received in the period. Replying to every review is rarely necessary, but replying to every review at three stars or below is a sensible floor. Replies appear publicly under the review, so the next customer reads your answer as much as the reviewer does.
Time to first reply is the gap between a review and your first answer, taken as a median so one forgotten review does not distort it. A target such as two working days is your own choice; track whether you meet it. A quick reply that is defensive or reveals private details does more harm than a slower, careful one.
Open issue age applies to everything in your register: a removal request awaiting a response, a correction asked of a publisher, a wrong AI claim, a listing to update. Log the date opened and closed, then read two numbers: how many are open and how old the oldest is. An issue open for ninety days with no recorded action is a process failure. Write the next action and its date beside each item.
Response metrics reward activity, which is not the objective: a perfect reply rate to reviews about a fault you have not fixed leaves the problem in place.
Related readWhat to do about negative content about me: triageAI claim accuracy and sentiment: count claims, not mentions
For AI assistants the unit is a claim, such as a founding year or a role. Ask the same fixed question several times and record whether each answer states the claim correctly, incorrectly or not at all. The metric is the share of runs that state it correctly. Repeated runs matter because an assistant can answer the same question differently within minutes. The full sampling routine is its own subject; the scorecard needs one line per claim per month: the accuracy share and the runs behind it.
Treat sentiment the same way. Automated labels can misread sarcasm and context, so read about twenty by hand each month and note how many you would change. A tool you overrule in a quarter of cases gives a rough sense of volume, not opinion. For a person or small business, your page one labels are a better sentiment measure.
Metrics that look useful and mislead
Some numbers feel like measurement and carry no information. Drop them, or label them as context.
- A single blended reputation score. It cannot say which part moved, and its weights belong to whoever built it. If you want a composite, build it from the metrics above and publish the weights.
- Raw mention volume. Mentions rise with fame and with scandal, so a rise has no direction.
- Follower counts and a rank for one query. The first measures reach, the second a sample of one.
- Average position without the page. Search Console reports your own site's best position, which can look steady while a negative result takes the space above it.
- Reviews solicited in a burst. They inflate the count and risk the policy problems above.
Be equally careful when a metric is offered as proof that a service worked. A rise in the owned share after a firm started is a reading, not proof of cause, because results also move on their own. Compare before and after against a stated plan and record what else changed. Whether reputation management works at all depends on the problem; the scorecard only tells you whether the work is moving the number it was meant to move.
A measurement calendar: what to read, when, and what you can conclude
Reading daily creates false signals, because results shuffle and one day tells you nothing; reading yearly misses the change that mattered. Monthly readings with a quarterly judgment suit most people, with extra readings around events.
Related readManage your online reputation yourself: a DIY plan| When | What you read | What you can conclude | What you should not conclude |
|---|---|---|---|
| Day 0 | Full baseline: queries, ratings, issue register | Where you start | Anything about direction |
| Weeks 2 to 4 | Page one log, ratings | Whether day 0 was typical | That a fix worked |
| Monthly | All families, one page | Large moves and open issue ageing | A trend from one month |
| Quarterly | Month to month series, shares and exposure index | Whether the plan is moving the numbers | That a change was caused by one action |
| After an event | Extra reading within a few days | The immediate effect of a story, launch or removal | That the effect will last |
Pair the calendar with tripwires written before the numbers move. For example: "If negative exposure rises by 3 points for two months running, I escalate; if the average drops by 0.2 in a month, I read the new reviews that day; if any issue is open 60 days, I call its owner." These are examples, not standards. What matters is that you set them while calm.
If readings by hand are more than you can sustain, our reputation monitoring page describes a monthly service that can take them over.
Next step: a one-page scorecard for the first month
Build one spreadsheet with the page one log, the ratings snapshot and the issue register, fix the definitions, and take the first reading this week. Add Search Console only if you own a site, and AI claim lines only for the three or four facts you cannot afford to have wrong.
If a result about you is already causing damage, take the baseline reading first so you can see what changes, then act on the most serious item. A free self-check of page one is a quick way to start the log.
At Reputation Partners we hold a scorecard to the standard of the work: dated readings, fixed definitions and plain words about what a number does and does not show. Write down the decision each number is meant to inform. If no decision follows, delete the row.
Questions people ask
What is the single most important online reputation metric?
There is no single one, but if you must pick, use weighted negative exposure on page one for your name, because it combines how many harmful results exist with how visible they are. Read it next to the owned share of page one, so you can see whether better material is taking the space.
How often should I check my online reputation metrics?
Monthly for most people, with a quarterly review of the trend. Daily checks are mostly noise, since results shift for reasons unrelated to you. Add a reading within days of an event, such as a news story or a removal.
Can I measure my reputation with a single score?
You can build one for your own use, but it hides which component moved, and a vendor's score depends on weights its maker chose. Keep the underlying readings beside it, and never treat it as proof that your reputation improved.
Are star ratings enough to judge a business's reputation?
No. An average depends on the number of reviews behind it and on how the platform computes it. Google's help page says its score is the average of all ratings published on Google, while Trustpilot says it uses the number of reviews, their age and a weighting system. Track volume, distribution and velocity, and read recent negative reviews.
Do I need paid software to track online reputation metrics?
Not to start. A spreadsheet, a private browser window and a fixed set of queries cover page one exposure, ratings and response, and Search Console is free for sites you own. Paid tools help at scale but bring their own definitions, which you should test first.
How long before the numbers show that my work is paying off?
It depends on the work. Replies and corrections at the source can change a reading within weeks. Pushing down an accurate result usually takes months, and ratings move only as fast as genuine reviews arrive. Judge on a quarterly series and decide in advance what counts as progress.
