GEO, SEO, Paid, Instrumentation
Rank a broken tag against a budget shift in the same list. Instrumentation is the silo nobody else has, fed straight from the tracking audits, so a defect gets queued instead of filed as a report you read and forget.
Beacon
Recommendations that show their own hit rate
One feed for every silo, ordered against a plan you actually set. Each rule carries how often it has fired, been followed, and been right.
Key takeaway
Every tool ships recommendations. Almost none of them tell you whether their own advice worked. Beacon keeps a record per rule of how often it fired, how often you acted, and how often the outcome was a win, and it orders the feed by that record.
One feed
Four silos, one queue, one ranking
A tracking defect, a missing schema block, a keyword you rank fourth for and a campaign past its saturation point are four different problems in four different tools. Here they are four rows in the same list, competing on urgency and effort rather than on which tab you happened to open.
Ranked on urgency against effort
Filter to what your team can actually ship this week. Each row carries an urgency, an effort level and a category, and the list sorts on a score built from all three.
Three actions, one lifecycle
Keep a queue instead of a growing pile of unread advice. Plan, Done or Dismiss, with each row moving from suggested to to-do to verified and the stage visible in a status column.
Every row has its receipts
Check the reasoning before you act on it. One full page per recommendation shows what fired it, the evidence behind it and what happened afterwards, so nothing has to be taken on faith.
Strategy
Advice is worthless without a stated goal
Most recommendation engines optimise in a vacuum, so they will happily tell you to cut the channel that is building the audience you said you wanted. A strategy is the plan the weekly narrative reasons against, and it changes what gets recommended.
One active plan at a time
Set the goal the advice reasons against. A month, a quarter or a year with an objective of leads, revenue or brand, and activating a new plan archives the old one because two live plans is the same as none.
A target the system can actually score
Pick a target the pacing can compute, so it actually gets checked. Six metrics: conversions, paid revenue, conversion rate, sessions, blended paid ROAS, GEO mention rate. The list is fixed on purpose.
Budget, channels and who you are for
Set the budget and priority channels once. The plan reads the personas and competitors you already defined rather than asking you to restate them, and takes optional per-channel targets.
A weekly read against the plan
Open Monday knowing whether you are ahead or behind. A narrative each week over the period so far, paced against your target rather than reported in isolation, ending in what to move.
Governance
We keep score on our own advice
This is the part that should be uncomfortable to publish. Each rule carries a running count of how often it fired, how often it was followed, how often the follow-up concluded, and how often it won. A rule that keeps losing gets demoted in the ranking.
Fired, followed, concluded, won
Judge the advice by its record. Four counts per rule kept as a contract rather than a dashboard widget, which is the same refusal that drives the readiness ladder: publish the number that could embarrass us.
A rule can be retired
Watch bad rules get removed rather than quietly kept. A precision verdict sits alongside the counts, so demoting a rule that produces plausible advice with poor outcomes is an evidence decision rather than a taste one.
The verdict is comparable across silos
Compare a GEO win against a paid win and have it mean something. One verdict rule covers every silo, because ranking rules against each other is meaningless when a win means ten mention-rate points here and two positions there.
What runs underneath the feed
The machinery that produces a row. None of it needs you to open it, and all of it is inspectable when a recommendation surprises you.
Budget optimiser
Hill response curves per campaign, modelling saturation so doubling spend is not assumed to double revenue, and an allocation where the marginal return is level across campaigns. Recommendations only surface above a meaningful shift threshold.
Experiments
A concluded geo holdout writes a calibration factor, that factor scales the response curve, and the resulting spend shift arrives as a recommendation with the test in its audit trail. The designs are built and no customer test has concluded yet, which the incrementality page states plainly.
Where it is available
Beacon is included from the Optimize tier upward, which is the same tier the AI engine probes start at. Gating the consumer above its producer would guarantee it never had data to read.
Before you trust the recommendation
Beacon ranks what to change. Whether the change caused the result is a different question, and it is answered with a control group.
FAQ
Direct answers.
Two things. The first is that every rule keeps a public record of its own outcomes: how often it fired, how often you acted on it, how often the follow-up concluded, and how often the result was a win. That record orders the feed, so a rule that keeps being wrong sinks. The second is that the four silos share one queue and one verdict definition, so a tracking fix and a budget shift are comparable rather than living in separate tools with separate notions of success. Most engines can tell you what they recommend. Very few can tell you how often they have been right.
Four silos. GEO covers AI visibility work: crawl access, extractability, citation gaps, sentiment. SEO covers keyword and content opportunities. Paid covers spend allocation, saturation and creative. Instrumentation covers tracking defects, and it is fed directly by the Trust Center audits and investigations, which is the join most stacks do not have: the tool that finds your broken tag is usually not the tool that queues the fix.
A strategy is one active, time-bounded plan per property: an objective of leads, revenue or brand, a target metric with a value, a budget, priority channels, and the personas and competitors you already defined elsewhere. The weekly synthesis paces against it instead of narrating metrics in isolation. The target metric comes from a fixed list of six because the pacing has to be able to compute progress period-to-date from data that exists. Free text would let you set a goal that silently never gets scored. Paid revenue is named that way for the same reason: the only revenue figure the pacing can read is the paid platform summary, so a goal set against total revenue would be judged on the paid slice of it.
No. Every recommendation needs a human to plan it, and the three actions are Plan, Done and Dismiss. The optimiser computes an allocation and the feed proposes the move; nothing changes in an ad account without you. That is a deliberate line, not a missing feature: an engine that publishes its own hit rate is an engine that admits it is sometimes wrong, and something that is sometimes wrong should not have your card.
Optimize and above. It sits at the same tier as the AI engine probes because Beacon reads probe output, and gating a consumer above its producer means it runs with nothing to read. Measure gives you the measurement layer: first-party tracking, deduplicated attribution, AI visibility and SEO keyword intelligence. Optimize is where the system starts telling you what to do about it.
14-day free trial
See what the system would tell you to do
14-day free trial. Beacon is included from Optimize, with the strategy, the weekly read and the rule record it keeps on itself.