This one is for the content leads, the demand gen marketers, and anyone who has spent three months optimising for answer engines and still cannot tell their boss whether a single thing has changed.
What: Using Breeze Assistant to build the measurement system that tells you whether your AEO work is landing: a scored set of the questions your buyers actually ask, a repeatable citation scorecard across ChatGPT, Perplexity, Gemini and Google AI Overviews, and the HubSpot reports that track AI referral traffic and branded-search lift. The point is to turn “is this working” from a shrug into a number you can put in front of leadership.
Prompt of the week:
Answer engine optimisation is, right now, the one marketing discipline where everyone is doing the work and almost nobody can prove the result. Teams rewrite their intros, add FAQ blocks, ship schema, and then look up expecting to see something move, and nothing does, because there is no ranking to watch. Answer engines do not rank pages, they cite sources, and if you grade the work on Google positions you will conclude it is failing while your citations are quietly climbing somewhere you are not looking. The absence of an obvious number to point at is precisely why so much AEO effort goes unmeasured and, eventually, unfunded.
The metrics that actually matter are new, and none of them live in your SEO dashboard. Citation count: how often you are quoted. Share of voice: your slice of the citations for a topic against your competitors. Appearance rate: the percentage of relevant questions where you show up at all. Mention sentiment: whether you are named, described, or actively recommended, and whether the description is even correct. And AI referral traffic: the visits arriving from chat.openai.com and perplexity.ai, which HubSpot now tracks natively alongside your other sources. A rankings tool will show you none of these.
There is also a trap waiting for the teams who do start measuring. AI citations churn violently: depending on the engine, somewhere between forty and sixty per cent of the sources cited for a given query change from one month to the next. Without a baseline and a steady method, a drop looks like failure when it is really just noise, and a spike looks like a win when it is really just weather. You need a fixed set of questions, scored the same way every time, run on a regular cadence, before any trend line means a thing.
The method every serious practitioner lands on is sound and thoroughly unglamorous: take twenty to fifty of the questions your buyers genuinely ask, run them across the major engines every month, and record whether you appear and how prominently. It works. It also quietly dies in a half-finished spreadsheet about three weeks in, because nobody designed it properly or made it repeatable enough to survive a busy fortnight. This prompt builds the system so it survives: the question set drawn from your real buyers, the scoring rubric, the cadence, and the HubSpot reports that catch the lagging signals, so measurement becomes a habit rather than a good intention.
Prompt structure
Paste this into Breeze Assistant and make sure CRM data access is enabled in your AI settings so Breeze can reference your products, personas, deal history, competitors, and traffic sources:
Role: You are an AEO measurement analyst. You know that answer
engines do not rank pages, they cite sources, so you measure the
things that actually move, citation share, appearance rate, mention
sentiment and AI referral traffic, rather than the rankings an SEO
tool would show. You also know AI citations churn heavily month to
month, so you design baselines and repeatable methods that separate
real change from noise.
Task: Design a complete AEO measurement system for us. Produce a
scored prompt set built from the questions our buyers actually ask, a
scoring rubric and a cadence, the HubSpot reports that track the
lagging indicators, and a monthly citation scorecard with share of
voice against named competitors. The goal is to turn "is our AEO
working" from a guess into a number I can defend to leadership.
Context:
- Company: [COMPANY NAME]
- Industry: [INDUSTRY]
- HubSpot tier: [note Marketing Hub Pro+ if present, since native
AEO tooling and AI referral tracking depend on it]
- What we sell and to whom: [the categories a buyer would ask an
answer engine about]
- Named competitors we want share of voice against: [LIST]
- Answer engines that matter most to our buyers:
[ChatGPT / Perplexity / Gemini / Google AI Overviews / Copilot]
- AEO work already done, and roughly when:
[so the baseline can be dated against it, or "none yet"]
- Who reads the scorecard: [marketing lead / board / founder]
Design the following:
1. PROMPT SET DESIGN
Build the set of questions we will test, from our reality, not a
generic category template:
- Derive the questions buyers actually ask at each funnel stage
from our CRM signals: personas, common sales questions, and the
reasons deals are won and lost
- Cover the four high-value question types: definition ("what is
X"), comparison ("X versus Y"), best-for ("best tool for Z"),
and how-to
- Include both branded prompts (do the engines describe us, and
do they describe us correctly) and unbranded category prompts
(do we appear at all when nobody is named). The unbranded
prompts are the real test
- Size the set to 20 to 50 prompts: enough for signal, few enough
to run by hand every month
- Flag the highest-intent prompts, the bottom-of-funnel purchase-
decision questions, so they are never the ones dropped when
time is short
2. SCORING RUBRIC
Make the scoring objective and repeatable:
- A 0 to 3 score per prompt per engine: 0 absent, 1 named, 2
described, 3 recommended
- How to record which sources the engine cited, so every gap
becomes a content brief rather than a shrug
- How to capture sentiment: are we described accurately and
favourably, or named but described wrongly
- How the per-prompt scores roll up into an appearance rate and a
share-of-voice figure against the named competitors
3. CADENCE & BASELINE
Design the discipline that makes the numbers trustworthy:
- Establish a baseline now, and do not judge any trend until
enough months have passed to see past the churn. State how many
months that should be
- A fixed monthly run: same prompts, same engines, same scorer,
same week of the month
- The rule of thumb for telling a real change from noise, given
that 40 to 60 per cent of cited sources churn month to month
- A lightweight log so the method survives a change of owner
4. HUBSPOT LAGGING INDICATORS
Wire up the signals HubSpot can track automatically:
- AI referral traffic: the report to build for visits arriving
from chat.openai.com, perplexity.ai and other AI sources, now
tracked natively
- Branded search and direct traffic trend, as lagging proof that
citations are building the awareness that converts
- Optional CRM link: the lifecycle stage and quality of contacts
arriving from AI referral sources, to connect citations to
pipeline
- If HubSpot's native AEO tooling is available, how it complements
the manual prompt set rather than replacing it
5. THE SCORECARD
A single page, produced monthly:
- Appearance rate, share of voice against competitors, a
sentiment summary, AI referral traffic, the biggest gains and
losses, and the top three prompts we still do not appear for
- Written for whoever reads it: a marketing lead wants the trend,
a board wants share of voice and the pipeline link
Constraints:
- Measure citations and appearances, not rankings. Do not import SEO
rank metrics and present them as if they prove AEO
- Establish a baseline before drawing any conclusion. Given the heavy
monthly churn, never treat one month's movement as a trend
- Build the prompt set from our actual buyers and CRM data, not from
generic category questions a competitor would also guess at
- Be honest about what is manual. Where a metric can only be captured
by querying an engine by hand, say so plainly rather than implying
HubSpot reports it automatically
- Keep the whole system sustainable by one person in about an hour a
month, or it will not survive contact with a busy week
- If a figure you need is not visible from the current context, state:
"SIGNAL MISSING: [what needs checking manually]"
Output format:
### I. MEASUREMENT READINESS SUMMARY
{3-sentence overview: what we can measure today, the single biggest
blind spot, and an overall rating: NO BASELINE / BASELINING /
TRACKING}
### II. SCORED PROMPT SET
| Prompt | Type | Funnel Stage | Branded? | Priority |
### III. SCORING RUBRIC
{The 0 to 3 scale, how sentiment and cited sources are recorded, and
how scores roll into appearance rate and share of voice}
### IV. CADENCE & BASELINE PLAN
{Baseline window, the monthly run method, the real-versus-noise rule,
and the log}
### V. HUBSPOT REPORTS TO BUILD
| Metric | Source | HubSpot Report | What It Proves |
### VI. MONTHLY SCORECARD TEMPLATE
{The one-page layout, tuned to who reads it}
### VII. 30 / 60 / 90 DAY ROLLOUT PLAN
Why this prompt works, and how to adapt it
There is no shortage of advice on how to do AEO. There is almost none on how to tell whether your AEO did anything, and that asymmetry is the whole reason this prompt exists. The optimising is well covered; the proving is a void, and a discipline that cannot prove its results is a discipline that eventually loses its budget. What every practitioner arrives at the hard way is that you cannot manage what you have not baselined, and you cannot baseline in a category this noisy without a fixed method run patiently over time. Having Breeze design the measurement system in a single pass is what closes that gap, turning a good intention that dies in a spreadsheet into a habit that produces a number you can stand behind.
A few things to note about how it is constructed:
It measures citations, not rankings. The single most common AEO measurement mistake is reaching for the SEO dashboard, watching the rank positions, and concluding nothing is happening. Answer engines do not rank, they cite, so the prompt deliberately refuses rank metrics and builds everything around citation share, appearance rate, sentiment and AI referral traffic. It is a different scoreboard for a different game, and using the old one is how teams talk themselves out of work that is actually succeeding.
The unbranded prompts are the real test. Anyone gets described tolerably well when you type their own name into an engine. The question that actually matters to growth is whether you appear when nobody is named, when a buyer asks for the best tool for a job or how two approaches compare. The prompt insists on both branded and unbranded questions precisely because the unbranded ones are the honest measure of whether the answer engines consider you a credible source rather than just a company that exists.
The baseline is the whole point. Because forty to sixty per cent of cited sources change from month to month, a single reading tells you almost nothing. The prompt is built around establishing a baseline first and then holding the method steady, so that when a number moves you can tell whether it moved because your work landed or because the engines simply reshuffled that week. Teams that skip the baseline end up celebrating noise one month and panicking about it the next.
It is built from your buyers, not a category template. A generic prompt set ("what is a CRM", "best marketing software") measures a competition you were never really in. The prompt tells Breeze to derive the question set from your CRM: your personas, your real sales questions, the reasons your deals are won and lost. That is exactly the work a standalone AEO tracker cannot do, because it does not know your buyers, and it is where pointing Breeze at your own data earns its keep.
It respects the line between manual and automatic. Some of this HubSpot can track for you, notably AI referral traffic and branded-search trends. Some of it, the actual citation scoring, still requires a human to query an engine and read the answer, because no report can see inside a live ChatGPT response for you. The prompt is honest about which is which, so nobody builds a scorecard that quietly pretends the manual part is automated and then goes stale the moment attention drifts.
“SIGNAL MISSING” keeps it honest. The fastest way to discredit a measurement system is to fill it with numbers that were quietly invented. Breeze can see your CRM and your traffic sources, but it cannot see live answer-engine output, and it does not know your true citation rate until someone measures it. Where a figure would otherwise be guessed, the flag hands the gap back to a person, because an honest blank in month one is worth far more than a confident number nobody can stand behind.
Adapting it for your portal:
On Marketing Hub Pro+? If you have HubSpot's native AEO tooling, add: “We have HubSpot's native AEO features. Show me how the built-in brand-visibility and prompt tracking should work alongside my manual scored prompt set, so I use the native tool for breadth and the manual set for the specific bottom-of-funnel prompts that matter most.” The output pairs the two rather than duplicating them.
Just starting AEO? If you have not optimised anything yet, that is the best possible moment to measure. Add: “We are about to begin AEO work. Prioritise establishing a clean baseline now, before any optimisation, so we can prove the lift later, and tell me exactly what to capture this month so month six has something to compare against.” The system becomes a before-and-after rather than a guess.
Fighting specific named competitors? If the goal is to win share from a known rival, add: “Weight the whole system towards share of voice against [competitors]. Include their branded prompts in the set, and tell me which category questions they currently win that we should target first.” The scorecard becomes a competitive dashboard rather than a solo report.
Local or regional business? If geography matters to your buyers, add: “Our market is [region]. Include location-specific prompts and account for how answer engines handle local questions, and tell me whether our Google Business presence is helping or hurting our appearance rate.” The prompt set gains the local questions a national template would miss.
Tight on time? If an hour a month is optimistic, add: “Cut the set to the 15 highest-intent prompts and design the cadence as quarterly rather than monthly, while still being honest about how much slower the trend will become visible at that frequency.” The system shrinks to something that will actually get done.
Want a quarterly cadence? Save the output and re-run it 90 days later with: “Compare against the output from [DATE] and report on how appearance rate and share of voice have moved, which prompts we now win that we did not, which we have lost, and whether AI referral traffic has risen in step.” That turns the scorecard into a running trend rather than a snapshot.
Beyond the prompt:
A measurement system is only ever as good as the discipline behind it. Two habits decide whether this one earns its keep or quietly lapses: how you start it, and how patiently you read it.
The first is to baseline before you optimise. The instinct, once you start caring about AEO, is to fix pages immediately, which feels productive and quietly destroys your ability to prove the fixing worked. Capture the baseline first, even if it is a bad one, especially if it is a bad one, because a low starting score is the thing that makes a later rise defensible. A month of patience now buys you a year of credible reporting later.
Run the first month honestly, by hand. The citation scoring genuinely does mean sitting down and asking the engines your questions and reading the answers, and there is no shortcut that does not also introduce a lie into your data. It takes an hour. Do it properly once and the rhythm sets; fudge it once and the whole scorecard becomes fiction that everyone eventually learns to ignore.
Then wire the HubSpot reports so the lagging indicators track themselves. AI referral traffic and branded-search trend are the signals that keep working while you are busy, and they are the ones that connect citations to something a board cares about, which is pipeline. Set them up once and they compound quietly in the background between your manual runs.
Then hold the line on not judging too early. The hardest discipline in AEO measurement is doing nothing with the first month's number except writing it down. The churn is real, the temptation to react is strong, and the teams that win are the ones that let the baseline mature into a trend before they draw a single conclusion. Three steady months beats three reactive weeks every time.
None of this makes the AEO work itself any easier. What it does is make the work legible: to your boss, to the board, and to the part of you that just wants to know the effort is not vanishing into a void. In a channel this new and this noisy, being able to measure honestly is not a nice-to-have. It is the thing that decides whether you get to keep going, because nobody funds a bet they cannot see paying off.
