Predict Before You Peek: Calibrating on 2025's AI-Visibility Research

Marketing Academy · Field Work●Brand Strategy

All projects·From: Brand Building in the AI Era: Visibility When AI Answers Replace Search

CoreForecast· 35 minutes

Coinbase

Objective: Predict the size of two real, measured AI-visibility effects before revealing the actual research numbers, then use the gap between your guess and reality to calibrate how much brand-content investment actually moves AI citation.

You're presenting a content-budget shift to Coinbase's marketing leadership, moving spend from paid performance toward citation-worthy content. Before you present, you want to know if your own intuition about what actually moves AI visibility is calibrated.

Write down your prediction for each stat before revealing it, then compare, then translate the real numbers into one budget-shift recommendation.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Docs

FreeRecord predictions before reveal, then draft the budget recommendation

Free, and writing the prediction down before checking forces honest calibration instead of hindsight bias

The process

2 steps

Step 01 of 02

Statistics and quotations increase AI visibility

The lesson states that adding statistics to content increases AI visibility by a measured percentage, and adding direct quotations boosts it by a larger measured percentage, per 2025 research cited in the lesson.

Before looking back at the lesson: do you predict adding statistics or adding direct quotations produces the bigger AI-visibility lift, and by roughly how many percentage points each?

Google Docs— Open a blank Google Doc and write your two numeric predictions before scrolling back to the lesson.

Procedure

Write your predicted percentage lift for adding statistics to content
Write your predicted percentage lift for adding direct quotations to content
Only after both are written, check the lesson's actual figures (22% for statistics, 37% for quotations) and record the gap

Sample output

MY PREDICTION: statistics +15%, quotations +20%
ACTUAL (per lesson): statistics +22%, quotations +37%
GAP: underestimated both, and underestimated quotations by more than double my guess for statistics

Healthy

A prediction within roughly 10 points of the real figures, and correctly guessing which lever (quotations) is bigger.

Unhealthy

Guessing statistics as the bigger lever, which is the more common intuitive mistake since numbers feel more 'objective' than quotes.

What this means

Direct quotations outperforming raw statistics is counterintuitive if you assume AI models weight numeric precision most heavily. In practice, a quotable, attributable sentence is easier for a model to lift into an answer verbatim.

So what do I do about it?

Symptom	Action	Effort
Content briefs ask writers to 'add data' but never ask for a quotable, attributable line	Add a required 'quotable sentence' field to the content brief template, not just a stats requirement	5 min

YouYou can do this yourself, no engineering access required.

Step 02 of 02

Community content compounds LLM equity

The lesson's Move 3 argues that community-generated content, forum threads, a Slack group, a subreddit, compounds because every post becomes an ongoing training signal for the next model generation, unlike a one-time blog post.

Given that most marketers still allocate close to 70% of spend to performance channels per the lesson, write a two-sentence case for leadership on why a small, recurring community-content budget line is worth adding this quarter.

Google Docs— Continue in the same Google Doc, add a short recommendation section.

Procedure

State the current allocation problem in one sentence (most budget chases clicks that are shrinking as AI answers replace search)
State the community-content case in one sentence, tying it to the compounding mechanic from the lesson, not just 'engagement'

Sample output

Nearly 70% of our spend still targets clicks that a growing share of buyers never make. Investing even a small recurring budget in community, a moderated subreddit or user forum, creates human-authored mentions that compound with every future model training cycle, unlike a blog post that only exists once.

Healthy

The recommendation names the specific compounding mechanic (training signal for future models) rather than a generic 'community is good' argument.

Unhealthy

A recommendation that describes community content as good for 'engagement' or 'brand love' without connecting it to the AI-citation mechanism the lesson actually explains.

What this means

Leadership will fund what they understand the mechanism for. 'It compounds because every post is a future training signal' is a specific, defensible budget argument; 'community builds brand love' is not.

So what do I do about it?

Symptom	Action	Effort
Budget proposals for community content get deprioritized against performance channels with clearer immediate ROI	Reframe the pitch around the compounding training-signal mechanic, not soft engagement metrics	5 min

YouYou can do this yourself, no engineering access required.

Final deliverable

A recorded prediction-versus-actual comparison for the statistics/quotation lift, plus a two-sentence budget-shift recommendation grounded in the community-content compounding mechanic.

See a reference example

Sample output

ThredUp calibration doc (excerpt)

MY PREDICTION: statistics +25%, quotations +25%
ACTUAL: statistics +22%, quotations +37%
GAP: overestimated statistics slightly, significantly underestimated quotations

Budget case: Nearly 70% of spend still chases clicks fewer buyers make. A small recurring community-content line compounds because every user post becomes a training signal for the next model generation.

Success criteria

You're done when you can:

Prediction recorded in writing before the actual figures were checked
Gap between prediction and actual figures explicitly stated
Budget recommendation names the compounding training-signal mechanic, not a generic engagement argument