Measure & Decide

The last skill is judgment: how to measure a probabilistic target without fooling yourself, and how to decide which lever to pull. Then a mixed review of the whole course.

Why this, for you: the techniques only pay off if you can tell whether they worked — and GEO measurement is full of traps that make noise look like signal. This lesson hands you a measurement discipline and a one-glance decision table tying the course together.

SEO rank tracking works because results are deterministic. GEO measurement is not — engines generate probabilistic outputs on the fly, with no rank API and no impression counts.

1 Measure by sampling, with eyes open

All GEO data comes from repeated sampling: run a fixed prompt set across platforms on a cadence and track presence. Borrowed SEO metrics don't map — use GEO-native ones.

Metric	What it captures
Share of Model	% of AI responses where your brand appears for category queries
Citation Share of Voice	Your citations as a % of total category citations
Generative Position	Average rank when the engine outputs a list

Sample at least 20–30 prompts daily across platforms. Smaller budgets can't tell genuine change from session variance — citation presence drifts substantially month to month even with no content change.

The traps that fake signal

Attribution gap: a ChatGPT-discovered visit that lands days later shows as direct traffic. Model-update blindness: providers update silently, so a drop may be a weight change, not your content. Single-platform fixation: a ChatGPT win may show zero lift on Perplexity. Don't rewrite in panic — it can cost you SEO for no GEO gain.

2 The decision table

One symptom, one lever. This is the course on a single card.

Situation	Pull this lever	From
Cited on ChatGPT, absent on Perplexity	Build a separate, fresher page; don't reuse the same one	L02
Page ranks well but is never cited	Lead each H2 with a 40–60 word answer; split multi-topic pages	L03
Claims read vague ("significantly faster")	Swap qualifiers for numbers, dates, named sources	L04
Want training opt-out without losing citations	Disallow training bots, allow retrieval bots	L05
Published llms.txt, citations flat	Expected — it's navigation; add schema for citation lift	L05
Few earned mentions, low overall citation	Invest off-site (forums, press); backlinks are secondary	L01
Strong page, but the domain isn't the cited source	Build interconnected coverage; register a Wikidata entity	L06
Citation drop with no content change	Check Search Console first — index/crawl/schema, not content	L07
Weekly numbers swing wildly	Raise to 20–30 daily prompts; narrow broad prompts	L08

3 When GEO isn't worth it

Honesty closes the course. GEO carries real ongoing cost — structured data, freshness cycles, per-platform tracking. It underperforms for niche or low-query topics (small citation target regardless of quality), fast-moving black-box engines (techniques go stale), and credential-dependent verticals (legal/medical trust can't be structurally faked). Measure before and after — or you can't tell.

↪ Your win: the whole discipline, in hand

Measure by sampling — 20–30 prompts daily, GEO-native metrics; mind the attribution gap and silent model updates.
Match symptom to lever — the decision table maps each failure to the one technique that fixes it.
Earned mentions + extractable structure + honest specifics — the three things that compound.
Know when to skip GEO — niche, low-query, or credential-bound content may not earn the investment.

Mixed review — the whole course

Question 1 · measurementA reliable GEO sampling budget is at least…

Question 2 · L01 + L05You published llms.txt and citations stayed flat. The right read is…

Question 3 · the trapsA ChatGPT-discovered visit that lands days later usually appears in analytics as…

Question 4 · decisionA page ranks well but is never cited. Your first lever is…

Question 5 · spaced recall from Lesson 07A citation drop with no content change should send you first to…

That's the course. You can name the target (citation), the engines (four backends), the structural moves (answer-first, atomic, dense), the technical layer (llms.txt, schema, crawler policy), and the measurement discipline. Want the Python sampling loop, or a GEO audit checklist for one of your own pages?