PR Super Dashboard

Where we stand, in one sentence

More people are talking about us and we are getting more press than last quarter, but when buyers ask AI for a recommendation we are still named less often than two of our competitors.

Sample data. Twelve weeks to 26 August 2026.

The number we are accountable for

Positive share of voice

36%

Of every positive thing said about this category, just over a third of it is about us. That is the number this agency is paid to move.

12 weeks ago: 28% → now 36% +8 points

How it is built

Volume alone can be bought. Sentiment alone can be tiny and irrelevant. Positive share of voice only goes up when people are saying good things and enough of them are saying it.

The three places it comes from

3.0/5

Overall PR health. The average of the three scores below. Public and Media are both strong at 4. AI scored 2 on visibility and was then marked down to 1, because an AI assistant is already repeating a complaint about us unprompted. That single point is the clearest statement of where next quarter goes. How this is calculated.

P

Public

Are people talking about us?

✔ Good

4/5

Strong, ahead of most

12,400 posts

12 weeks ago: 9,900→ now 12,400+25%

People are posting about the brand a quarter more often than they were three months ago, and most of it is positive. Nothing here needs fixing.

How much of it is positive68%

Our share of category talk24%

Complaints needing a reply31

M

Media

Are we getting written about?

✔ Good

4/5

Strong, ahead of most

1,284 stories

12 weeks ago: 1,146→ now 1,284+12%

We are in the press more than last quarter, and a fifth of it is in the big four papers. The messages we want repeated are getting through about six times in ten.

In tier 1 outlets214

Our key messages picked up62%

Negative stories9

AI

Do AI assistants recommend us?

▲ Needs work

1/5

Poor, needs intervention

Scored 2 on visibility, marked down to 1: an assistant is already repeating a complaint about us

3 times in 10

12 weeks ago: 2 in 10→ now 3 in 10improving

Ask ChatGPT or Google AI to recommend a brand in our category and we get named about three times in ten. Two competitors are named more often than we are. This is the weakest of the three, and the one nobody is working on.

Where we rank3rd of 6

Named most oftenMeridian

News as a source for AI2.8%

5 best in category 4 strong, ahead of most 3 fair, middle of the pack 2 weak, behind competitors 1 poor, needs intervention

✔ Good no action needed ● Watch keep an eye on it ▲ Needs work decide something

Where these scores come from. Every band is set against the competitive field rather than a target we chose, so a score can only improve by taking share from someone. See the full methodology, including the modifiers that pull a score down.

Radar: what is growing that we should worry about

Issues rising fastest right now

Ranked by how far each one is above its own normal level, not by how big it is and not by how fast it is growing. Some themes are simply noisy, so a doubling is ordinary for them. Others barely move, so a small rise is genuinely strange. Only the second kind is news.

Open the full radar, including the ones that are big but not moving.

What changed, in plain English

✔
The technician tour was the best thing we did all quarter. It cost ₱900,000 and produced 96 stories, which works out at about ₱11,000 per point of score. See the full ranking.
▲
The Q3 launch event cost nine times more than the technician tour and scored less than half as well. Worth asking whether we run it again.
●
Two of our seven key messages are barely appearing in coverage at all. "Service within 48 hours" reached 19% of stories and "Filipino design team" reached 6%. See which messages landed.
▲
When AI assistants answer questions about our category, only 2.8% of what they read is news coverage. They mostly read Lazada, Shopee, spec sites and Reddit. Press releases alone will not move this. See what AI is actually reading.

What that suggests we do

1
Run the technician tour again next quarter. It is the cheapest score we have.
2
Put the two dead messages in front of the client. Either drop them or change how we brief them in.
3
Because AI reads retail listings and forums rather than press, some of the PR budget probably has to go into retail product content and community seeding. That is a conversation to have, not a decision for us alone.

If you do want the chart

How often each AI assistant names us, week by week

Higher is better. All four lines have been climbing since June.

Weighted score by execution

Coverage, social lift and share-of-model movement, combined with the weights set in the scoring config. Higher is better.

The detail

Execution	Type	Pillar	Score	Coverage	Tier 1	Spend	Cost per point

Efficiency, not just volume

Best value

Aircon technician tour

₱11k per point of score

Worst value

Q3 launch event

₱214k per point of score

Median

₱48k per point

across 18 executions

Why this page exists

A crisis is almost never a surprise. It is a small complaint behaving unusually for six weeks while everyone watched the totals go up.

Every theme is measured against its own history, so the question is never "is this big" but "is this normal for this".

How it decides something is wrong

Each theme is judged against its own history, not against the others

The biggest complaint in the list is usually not the problem. It is usually just the thing people always complain about.

Step 1

Learn what normal looks like

For each theme we hold the level it usually runs at and how much it usually bounces around, built from its own trailing weeks.

Step 2

Measure the gap

A theme running well above its own usual variation is flagged. One running high but within its usual variation is not, no matter how large.

Step 3

Improve the baseline

Every week adds to each theme's history, and the thresholds get re-derived from problems that actually escalated. False alarms fall on their own.

Two deliberate brakes. A theme under 25 mentions is never flagged however extreme the maths says it is, because three complaints where there is usually one is not a crisis. And a theme with fewer than eight weeks of history is marked provisional baseline, because there is not yet enough of its past to say what normal means.

The scope

How to read it

Four things are being shown at once, and none of them is decoration.

Closer to the middle means further from its own normal level

Which third of the dial shows how far it has spread: public talk, then news, then AI answers

Bigger dot means more people are saying it

Red and pulsing means it is escalating now

A dial is good at getting attention and poor at precise comparison. Use it to see what has moved, then read the quadrant and the list below for the actual numbers.

How an issue escalates

Stage 1

People complain to each other

It sits in social and forums. Cheap to fix here. Usually a service or communication problem, not a product one.

Stage 2

A journalist picks it up

The moment a theme crosses from social into news, the cost of handling it jumps. This is the crossing the radar watches for.

Stage 3

The AI starts repeating it

Once a complaint is established enough that AI assistants mention it unprompted, it repeats to every buyer who asks, indefinitely. A news cycle ends. This does not.

Everything on the radar

Big versus abnormal

Left to right is how much people are saying it. Bottom to top is how far it is above its own usual level. The dotted line is the only threshold that matters: anything below it is behaving as it always does, however large it is.

The list

What this page cannot do

It finds themes that are growing in the sources we watch. It will miss anything happening somewhere we do not collect, anything in a private group, and anything the sentiment model reads as neutral because it was phrased politely. Treat a quiet radar as "nothing found", never as "nothing there". Every number here should be clickable through to the actual posts and articles behind it before anyone acts on it.

Positive share of voice

36%

Of all the positive things said about this category across news, social and AI answers, 36% of it is about us. Up 8 points in twelve weeks.

Positive share of voice, by brand

This is share of positive talk, not share of all talk. A competitor can be talked about more than us and still sit below us here.

Sentiment split, us against the field

Centred on neutral. Bars to the right are positive, to the left negative. Net figures hide crises, so the negative half is always shown at full size.

Positive Neutral Negative

Where the positive talk is coming from

Public conversation

68%

▲ 5 pts, positive

News coverage

41%

▲ 3 pts, positive

AI answers

72%

of answers naming us are favourable

Which executions actually created positive sentiment

Net positive mentions generated, per peso spent

This is the ranking that matches what we promise. Volume-only ranking is on the leaderboard page and it puts a different execution first.

Read this before quoting the number to a client

Sentiment is the least reliable thing in this stack. It is a judgement made by a model about tone, and it gets sarcasm, mixed reviews and Taglish wrong more often than it gets volume wrong. Three rules follow from that. Always show the negative bar at full size rather than only a net score. Always be able to click through to the actual articles behind a number. And never let positivity be the only thing an execution is judged on, or the incentive becomes soft coverage nobody reads.

Share of voice

Every mention of any brand in this category, across news and social. Ours is highlighted; competitors are shown in grey so the comparison is with us, not between them.

Brand	Mentions	Share	Change

The same picture, positive mentions only

Positive share of voice

The one above counts every mention. This one counts only the favourable ones, which is a different ranking and the one we hold ourselves to.

Brand	Positive mentions	Share of positive	Versus all mentions

Share of model over time

Each engine is asked the same panel of buyer questions every week, three times, to smooth out randomness. The line is the share of answers that name the brand.

By message pillar

Where the brand is visible, and where it is not

Same measure, split by the pillar each question belongs to.

Who gets named instead

Competitors appearing in the same answers

Share of answers naming each brand, whether or not ours appears.

2.8%

Earned media is a rounding error in what AI cites

This is the number that should change how budget is split. The engines are overwhelmingly reading retail listings, spec sites and forums. Press coverage is present, but barely.

Where AI answers get their sources

1,808 citations across 800 answers. One measure, so one colour.

The individual sources

Source	Type	Times cited	Share	Can PR influence it?

Coverage volume with executions marked

Bars are articles captured per week. Markers show when a PR execution went out, so a lift can be attributed rather than guessed at.

Against the prior baseline

Lift versus the 7 days before each execution

Positive means the week after beat the week before.

Which key messages actually made it into coverage

Share of brand articles in which each approved message was detected.

By outlet tier

Message	Tier 1	Tier 2	Tier 3	Total	Pickup rate

All outlets Manila Bulletin Inquirer PhilStar Rappler

Everything Mentions the brand Mentions a competitor

Headline	Outlet	Published	Pillar	Mentions

What this is

One dashboard that pulls together public conversation, earned media and AI answers, scores each out of five, and ranks PR work by what it actually produced rather than by how much of it there was.

Prototype with sample data. Twelve weeks to 26 August 2026.

The three scores

Every band is set against the competition, not against a target we chose

This is the part worth defending in a client meeting. A target we set ourselves is one we chose, and a client is right to discount it. A position against the field is not.

Fair share. If six brands are tracked, an average brand would hold a sixth of the talk, so 16.7% is fair share. The index is what we actually hold divided by that. An index of 1.0 means we are performing like an average brand in this category.

Because the index is relative, the same bands work for a client with three competitors and one with twelve, with no retuning. And a score can only be improved by taking share from somebody else, which is what the client is paying for.

Score	Public and Media	AI	Reads as
5	2.5x fair share or better	Named as often as the leading brand	Best in category
4	1.75x to 2.5x	75% to 95% as often as the leader	Strong, ahead of most
3	1.0x to 1.75x	60% to 75% as often	Fair, middle of the pack
2	0.6x to 1.0x	40% to 60% as often	Weak, behind competitors
1	Below 0.6x	Named less than 40% as often	Poor, needs intervention

Why each score is measured the way it is

Public

Measured as our share of all positive mentions in social listening, not our share of all mentions. If the agency sells positive sentiment as the return, the score has to be denominated in the thing being sold. Otherwise a quarter where volume rose and goodwill fell still scores well.

Media

Same measure, applied to earned coverage. Tier and message pickup are handled as modifiers rather than blended into the headline number, so the score stays explainable in one sentence.

AI

Measured against the leading brand rather than as a share of a total. An assistant can name six brands in one answer and one in the next, so share of a total moves for reasons that have nothing to do with us. How often we are named next to whoever is named most is stable.

Modifiers

What stops a headline number from flattering us

A share number can look healthy while the thing underneath rots. High share of positive talk alongside a growing complaint pile is a real state, and a dashboard that scores it a 4 is one that stops being trusted. So four conditions pull a score down and one pushes it up.

-1
More than 15% of our own mentions are hostile.
-1
An issue is escalating in that layer, per the issue radar.
-1
Less than 10% of our coverage is in tier 1 outlets. Volume in places nobody reads is not coverage.
-1
An AI assistant is already repeating a complaint about us unprompted.
+1
More than 30% of coverage is tier 1 and message pickup is above half.

Nothing can go below 1 or above 5. The overall score is a plain average of the three, deliberately: a weighted one invites an argument about the weights every time somebody dislikes the result, and there is no evidence yet for claiming one layer matters more than another. Revisit after a year of data.

This period, worked through

Score	Raw measure	Index	Band	Modifier applied	Final
Public	35.8% of positive talk, fair share 16.7%	2.15	4	none	4
Media	35.8% of positive talk, fair share 16.7%	2.15	4	none	4
AI	Named in 31% of answers, leader 54%	0.57	2	-1, a complaint is already repeating in AI answers	1

Overall is 3.0. It would read 3.3 without the modifier. We show the lower number, because a scoring system that quietly rounds in our favour is worth nothing the first time a client checks it.

Where the data comes from

Layer	Source	How often	Status
Media	Live RSS from the major Philippine outlets	Every 30 minutes	Live
Media	Coverage exports from the client's monitoring tool	On upload	To build
Public	Exports from the client's social listening tool	On upload	To build
AI	A fixed panel of buyer questions put to the AI engines	Weekly	Exists, to connect

How this gets better over time

Nothing here learns yet. It remembers, which is the part that has to come first

You cannot learn from data you did not keep. Every week this runs without recording the right things is a week of training data thrown away, so the first work was not a model, it was memory. Four things happen in order, and only the last one looks like machine learning.

First

Remember

Every metric is written to an append only history, together with the numbers behind it and which version of the rules produced it. That is what makes it possible to ask "would last quarter have scored differently under the new rules". Live now.

Second

Be corrected

When someone says a sentiment call is wrong, or an article is not about this client, that disagreement is stored with who said it and why. This is the most valuable data in the project and it is produced for free, but only if there is somewhere to put it.

Third

Be scored

The radar makes a prediction every week. Both the prediction and what actually happened are written down, so it becomes possible to say how often the radar is right. A warning system nobody has measured is one nobody should trust.

Then

Be calibrated

Hand set thresholds get replaced by what has actually been observed. A 5 becomes the top of the real distribution rather than a number somebody chose. The radar's alarm level gets set by what really did escalate.

The radar is the clearest case. Right now each theme's normal level comes from its own trailing weeks. That baseline improves by itself: every week adds history, seasonal patterns become visible, and themes that always spike stop triggering alarms. Nobody has to change any code for that to happen.

Two rules that keep it honest. A human correction always beats a model output, permanently, and is never deleted, only superseded. And no recalibration takes effect on its own: a proposed change is replayed over all of history first, and somebody has to accept it. A system that silently moves its own goalposts is not learning, it is drifting.

What this cannot do

Sentiment is the least reliable thing here. It is a model's judgement about tone, and it gets sarcasm, mixed reviews and Taglish wrong far more often than it gets counting wrong. Every sentiment number should be clickable through to the posts behind it before anyone acts on it.

The radar only sees where we look. It will miss anything in a private group, anything on a platform we do not collect, and anything phrased politely enough to read as neutral. A quiet radar means nothing found, never nothing there.

Attribution is a judgement, not a fact. Coverage is linked to an execution by date window and keyword match. It will sometimes credit the wrong thing. Manual overrides are kept and never overwritten by the automation.

Nothing here proves sales. These are measures of visibility and sentiment. They are leading indicators at best, and should be presented that way.

Words we use

Term	What it means
Share of voice	Our share of every mention of any brand in the category
Positive share of voice	Our share of only the favourable mentions. The number the agency is accountable for
Share of model	How often an AI assistant names us when asked for a recommendation in our category
Fair share	What one brand would hold if the category were split evenly. 100 divided by the number of brands tracked
Tier 1	The outlets that move opinion. Set per client in the outlet config, not by us
Message pickup	The share of our coverage in which an approved key message was actually detected
Execution	One piece of PR work. A release, a pitch, an event, a byline, an influencer send
Escalating	An issue growing fast enough, or spread far enough, to need a decision this week

Built by Ripple8 Insights. The scoring bands live in a settings file so they can be changed without touching the system, and this page is generated from the same definitions the scores use. If the two ever disagree, the settings file is right and this page is stale.