01 — Start here
Three things it found in its first week of our own data
Search Console access landed on 30 July. These are real numbers from the last 28 days, not a demo. If you only read one section, read this one — it's what the engine is for.
We rank #2 for "seamless customer service" and get nothing
22,029 people saw us in position 2. Not one clicked. Either an AI Overview is eating the answer above us, or the title and description are simply not clickable. One look at that live search result tells us which — and that answer changes what we do next across hundreds of pages.
source · Search Console, 28 days to 27 JulThe glossary is outranking the pages that sell
For "call center software", our glossary entry and our solutions page are competing with each other at almost identical positions. Same story for "contact center automation", where two different glossary entries fight over the same term. We're splitting our own strength.
source · Search Console, query-to-page mappingThe market renamed what we sell
Searches putting us in front of people for "ai call center" grew more than five-fold in a month, while "performance management systems" fell 79%. Same product, different words. Our page titles and hubs are still largely written in the falling language.
source · Search Console, 28 days vs prior 28Until last week the engine was reading the market from the outside — news, and a third-party estimate of where we rank. Search Console is the first source that reports what Google actually showed people, and what they actually did about it. It's the difference between an opinion about our visibility and a measurement of it.
02 — The point of the whole thing
It doesn't find keywords. It answers six questions.
That distinction is the founding idea and it's worth being pedantic about, because it's what makes this different from every keyword tool we already have.
| # | The question | What you get instead of a keyword list |
|---|---|---|
| 1 | What topic is emerging? | A ranked shortlist, not a firehose — scored 0–100 on relevance, demand, competitor presence and newsworthiness |
| 2 | Where is it showing up? | Which surface it's moving on: Google, news, AI answers, or conversation |
| 3 | What format is winning? | "Don't write another standard blog" — the format the results page actually rewards |
| 4 | Which team should act? | One named owner per opportunity. No opportunity enters execution without one |
| 5 | What should they make first? | A drafted brief: angle, hook, target keywords, a publish-ready post, a call to action |
| 6 | Amplify with paid or PR? | Only after organic traction, strategic importance, or genuine time-sensitivity |
The table above describes the design. This is what it actually produced, on 5 August 2026, for a real signal — and exactly where to look at each step. Nothing below is illustrative; every value is copied from the run.
| # | The question | What it answered, for real | Where you look |
|---|---|---|---|
| 1 | What topic is emerging? | “Conversational AI / chatbot search demand cooling” — scored 63, so it entered the weekly queue rather than act-now | Opportunity Board sheet, tab Opportunity Board — sorted by Score, highest first |
| 2 | Where is it showing up? | Search demand. SEMrush volume for chatbot and conversational ai is falling while agentic-AI phrasing rises |
Board column Trigger Type; underlying figures in Supabase semrush_keywords |
| 3 | What format is winning? | Named in the brief under Formats — a LinkedIn post plus a positioning piece, not another standard blog | Content Briefs Doc — the Formats heading in each block |
| 4 | Which team should act? | Owner assigned, priority this-week. No opportunity reaches a brief without one | Board columns Primary Team / Supporting Teams; also in the brief header |
| 5 | What should they make first? | A written brief, BRIEF-20260805-6EB93316, with the angle: “the word ‘chatbot’ is dying in search because the category graduated — buyers now search for agentic AI outcomes”, plus a publish-ready LinkedIn post |
Content Briefs Doc — newest brief is at the top; the title is coral when the priority is urgent, blue otherwise |
| 6 | Amplify with paid or PR? | Not yet — and that is the honest answer. This gate opens only after organic traction is measured, and outcome tracking has no published article to measure yet | Supabase content_publications — currently the thinnest table in the system |
1. The Opportunity Board sheet — every scored opportunity, highest first. Rebuilt from the database every morning, so it is always the current picture rather than a log. A Needs Review column flags any opportunity whose sources could not be resolved.
2. The Content Briefs Doc — the written briefs, newest at the top, colour-coded by urgency.
3. The daily digest — only what scores 60+, so it stays short.
If a number on the board looks wrong, that is worth reporting: the engine checks itself fourteen ways after every run and reconciles its click totals against Google’s own figures, currently at 102.8%. It is designed to be checkable, and it has been wrong before.
Which alerts are yours
Every opportunity is tagged with a trigger type, and the type decides the owner. Two of the five land on your desks; the other three go elsewhere. This is so nobody has to read everything.
| Trigger | Fires when | Owner |
|---|---|---|
| Search Demand | Volume rising, rankings moving, a competitor enters the results page, an AI Overview appears | SEO — yours |
| Format Gap | The results page is dominated by video or comparison pages and we've covered the topic in the wrong format | SEO + Creative — yours |
| Conversation | A topic climbs on Reddit or LinkedIn; a recurring customer pain surfaces | Social |
| PR / Narrative | A macro trend in the news, a competitor CEO quoted everywhere, an earnings or regulation moment | PR / Comms |
| Customer Signal | Several customers hit the same challenge; the pain shows up in sales notes | Customer Marketing |
Everything scoring 40 or above is written to the board. 60+ reaches the daily digest. 80+ is flagged act-now, 60–79 weekly queue, 40–59 monitor only. The maths is fixed and repeatable — the same signal on a Monday and a Thursday scores the same, so we can later check whether the scores were any good.
03 — Coverage
What's actually connected, honestly
"Live" means it runs on its own and has produced real output. Nothing here is aspirational, and one row below is genuinely bad news.
| What it watches | The question it answers for you | Status | Note |
|---|---|---|---|
| News & competitor moves | What's the market talking about this week? | live | Daily, our own and rivals' announcements |
| Search demand & rankings (SEMrush) | Which terms are growing, which are dying? | live | Since 24 Jul, 12 tracked terms |
| Our own Search Console | What did Google actually show, and did anyone click? | connected, not flowing | Access landed 30 Jul; the daily feed still has to be built |
| AI answer engines | Do ChatGPT, Claude, Gemini, Perplexity mention us? | measured, not scheduled | Four engines tracked; last reading is stale |
| The brief writer | What do we make first, and what does it say? | live | 11 briefs written, grounded in live web search |
| The results checker | Did what we published actually move anything? | live | 21-day cycle — see §05 |
| Reddit & LinkedIn conversation | What are buyers saying before they search? | blocked | Needs an account approval |
| Sales win/loss reasons | Why do we actually lose deals? | not started | No agreed source yet |
| The engine itself | Is it running right now? | paused 6 days | Last run 24 Jul — it lives on a laptop, not a server |
The engine has not run since 24 July. It works — it has produced real briefs and real findings — but it sits on a local machine rather than a server, so it only runs when that machine is on. Everything else on this page is real; none of it arrives on your desk every morning until that changes. It's the same unresolved decision from the July briefing, and it's question 1 at the bottom.
Why the earlier reports said "live" and this one says "paused"
Both were true when written. Through July the engine ran daily on a local install and produced the 11 briefs and the scored board. It stopped when work moved to connecting new data sources — Search Console access, the search-demand feed — rather than running the existing loop. It is a pause, not a failure: nothing is broken, it just isn't scheduled anywhere permanent.
The honest reading: we have spent July proving each piece works, and have not yet spent a week letting the whole thing run unattended. That's the next milestone, and it's cheap once the delivery channel is chosen.
04 — What it does for you
Four jobs, at four different speeds
These aren't equally mature. Two run today, one just became possible, one needs the conversation feed unblocked. Each card says which.
Check whether it worked running
The most concrete of the four. Every published piece gets re-checked after three weeks — did it reach the results page, did our share of AI answers move more than a competitor's, did anything start citing us. A flat verdict doesn't get filed away; it comes back as a Refresh job.
Spot a trend early running
News and competitor announcements move weeks before search volume does. When a rival launches something, or an analyst piece lands, it's scored and on the board the same day — while there's still time to have an opinion first rather than a response later.
Predict what's coming partly
Rising and falling demand is measured week over week rather than guessed — so "ai call center" showing up as a five-fold riser is a fact with a date on it. What's still missing is the tie-in to 3,999 tracked terms sitting unused in our SEMrush account, including the 848 that already trigger an AI Overview.
See what we're missing new this week
The newest and least proven job, and possibly the most valuable. Put market demand next to our own impressions: if the market searches heavily for something and we appear for almost none of it, we're missing the trend. If we appear everywhere and nobody clicks, we're present but invisible.
None of these four decide anything. The engine narrows a very wide field to a short list, attaches the evidence and the reasoning, and drafts a starting point. A person still picks what ships, checks the facts, and owns the result. What it removes is the blank page and the hour of scanning before it — not the judgement.
It also cannot see anything it isn't pointed at. It missed our own Cognigy story entirely for three weeks — 146,824 people saw us for Cognigy searches and the engine never noticed, because nothing told it to look. That's fixed by adding a term to a list, but it's a good illustration: this is a very fast reader, not an oracle.
05 — The bit that closes
Publish, wait 21 days, then answer for it
Most content operations stop at "published". This one treats publishing as a bet, and comes back to settle it. Here's the whole cycle.
And then the important arrow: a flat verdict becomes a new Refresh job on the board — the failed bet re-enters at step 1.
What "moved" actually means
Three checks, and any one of them counts:
| Check | What it asks | Why it's fair |
|---|---|---|
| Did we show up? | Is the published page in the results for its target terms — and who else is there? | The most direct test, and the fastest to read |
| Did our share of AI answers grow? | Across four AI engines, did our share rise more than the top competitor's? | Comparing against a rival means a good month for the whole market doesn't get counted as our win |
| Did anything start citing us? | Did the new page begin earning citations for its question group, and on which engine? | Catches wins that never show as a ranking — increasingly where buyers actually are |
Three weeks is long enough for indexing and for AI answers to re-crawl, short enough that a failed bet doesn't sit unexamined for a quarter. It's a default, not a rule — a news-reactive piece can be checked in a week, a pillar page given six. Whoever pastes the URL can set the window.
The longer-term payoff nobody talks about yet
Right now the scoring weights — how much relevance matters versus competitor presence versus newsworthiness — are principled guesses. They're consistent guesses, which is the important part.
Once roughly ten published pieces have been through a 21-day check, we can compare what the engine predicted against what actually happened, and replace the guessed weights with observed ones. That is only possible because the scores are fixed maths and every piece ends up with an outcome label attached. It's the difference between a system that has opinions and one that learns.
06 — Why this should pay off
Four reasons, each with something to check it against
| The reason | What it changes day to day | How we'll know it's true |
|---|---|---|
| Effort goes where evidence points | A ranked shortlist replaces "what should we write this month?" — with the reason attached to each item | Briefs that came from a high score should move more often than ones that didn't. The 21-day verdicts will show it either way |
| Fixing beats creating, sometimes | The clearest wins this week weren't new content — they were a broken click rate and a glossary competing with itself | 25 terms we already rank for are worth roughly +667 clicks a month if they reach position 5. Small, and that's the point: it says fix the click rate first |
| Nothing is declared a success by default | Publishing is a bet that gets settled, not a task that gets closed | Every flat verdict produces a Refresh job. Count them — if that number is zero, the check isn't working |
| It gets sharper as it runs | Every source added feeds the same scoring, and every verdict is training data for the weights | After ~10 reviewed pieces the guessed weights can be replaced with measured ones |
Half our search visibility is on-topic and earns almost nothing: two million impressions for genuinely relevant terms produced 1,934 clicks in 28 days. Meanwhile off-topic pages — reverse phone lookup, proxy servers — earn more than every contact-centre term combined.
No amount of new content fixes that. It's a click-through and consolidation problem, and it was invisible to us until last week. The engine's first real contribution is arguably telling us to stop doing the thing we were about to do more of.
07 — Not there yet
What's missing, and what each one would buy
Ordered by what unlocks the most. None of these are hard engineering; most are a decision or an approval.
| Missing | What we lose without it | What it needs | Owner |
|---|---|---|---|
| 1 A permanent home | Nothing arrives on your desk automatically — the engine only runs when a laptop is on | A delivery channel decision, then the move to a shared server | You + AP |
| 2 Search Console flowing daily | The findings in §01 are a one-off snapshot, not a monitor. We also lose history daily — Google only keeps 16 months and the oldest day drops off every day | Build the daily feed. Access is already granted | AP |
| 3 3,999 tracked keywords | Position movement, and 848 terms that trigger an AI Overview, sit unread in a subscription we already pay for | Connect the existing campaign. No new spend | AP |
| 4 Conversation listening | We see the market only after it reaches news or search — not while buyers are still asking each other | An account approval | Social |
| 5 Why we lose deals | Content answers the objections we assume buyers have, not the ones that actually cost us deals | Agree a source with sales leadership | Sales + AP |
| 6 A weekly rhythm for AI answers | Our AI-answer share is measured but read too rarely to trust — the latest reading is weeks old | Decide a run cadence | AP |
Where we've measured it, NiCE is cited 24 times on Perplexity but only 3 times each on Claude, Gemini and ChatGPT. That gap is itself an opportunity — three of the four engines a buyer might ask barely know us. It's not on this list as a missing capability because we can already see it; it's here because nobody currently owns doing anything about it.
08 — Open questions
Four decisions, each with a recommended default
These are all of them — nothing else in this document is waiting on you. If the defaults look right, "approve with defaults" is a single reply and everything above moves.