GEO-111·Issue 10·Strategy & Market·111 / 132
A GEO Budget Model: Work Backwards from Traffic Value
Budget GEO from what diverted traffic is worth, not article count: four sourced inputs, one ceiling formula, five payback-ranked cost layers, three tiers.
Issue focusMeasurable: AI Visibility, Attribution, and Experiments
A November budget review at a DTC espresso brand selling into the US and Germany. The growth lead presents next year’s GEO plan: 120 articles, three monitoring seats, a part-time writer. The CFO asks how much gross profit it returns and when. Silence, then the plan is reclassified as “brand investment” and cut in half.
A GEO budget should not be built from article counts and tool seats; it should be reverse-engineered from what diverted traffic is worth and how fast it returns, which decides the layer you fund first and the pages you can afford.
Why “price per article” loses the room
Article count is a cost unit: it answers what you spend, never what comes back. The deeper problem is the missing anchor.
SEO budgets had one — search volume, cost per click, the organic traffic value column — so a CFO could price a ranking. AI answers publish no volume, guarantee no click, and name you today and a rival tomorrow; spend rests on conviction, which finance prices near zero.
Timing compounds it: crawl fixes are paid now, returns arrive after a recrawl, and off-site evidence cannot be clawed back. Finance counts payback within a year; content talks about an asset. The line gets cut.
Argue in finance’s language: ceiling, payback, capacity. Capacity was never articles; its unit is the answer unit, one self-contained claim a model can lift. Pricing by article teaches a team to pad the count.
Four inputs, one ceiling
Working backwards turns a hunch into four parameters you can argue about and source.
| Input | Source | Common error |
|---|---|---|
| Affected sessions: organic traffic on the questions you sell into | Search console, site search, support tickets | Using total site traffic |
| AI-read share: sessions that read an AI answer first | Sampled prompts, referral share | Assuming everything is replaced |
| Conversion and order value on AI-referred sessions | Attribution, order data | Applying the sitewide average |
| Gross margin after product cost, shipping, returns | Finance’s definition | Reporting GMV as the return |
Multiply them for a monthly ceiling: affected sessions × AI-read share × conversion × order value × gross margin, times twelve. That ceiling is not a forecast; every budget line must fit under it.
| Step | Action | Passes when |
|---|---|---|
| 1 Frame the question set | 20 category questions from tickets and prompts, each with a session range | Every question traces to a source |
| 2 Estimate share and conversion | Back out from referrals and attribution | A floor and a ceiling are written |
| 3 Restate as net margin | Strip returns and fulfillment from every number | One margin definition, no GMV |
| 4 Set the band | Take 30–50% of the ceiling, then split by tier | The band survives a 2× error |
Build a range, not a point: if the conservative ceiling cannot cover one hire, do not fund the program.
Example: the ceiling on a $280 product
Example — brand, volumes, and prices are invented for the arithmetic.
Twenty core questions carry 42,000 organic visits a month; sampling suggests 25% read an AI answer first. AI-referred sessions convert at 3.5% on a $280 average order, with 32% margin after returns.
That is 10,500 affected sessions, roughly 368 orders, about $103,000 in monthly revenue and $33,000 in gross profit. Annualized, the ceiling nears $396,000; a 30–50% band gives roughly $120,000–$200,000 a year.
At a 15% read share the band falls to $70,000–$120,000, and the meeting shifts from whether to do this to which assumption to run.
Sort the 20 questions by sessions × margin: the top five carry most of the value; funding those beats 120 articles.
Five buckets, five payback clocks
Crawl fixes and third-party reviews return at different speeds. An unlayered budget cuts the slowest bucket first, often the irreplaceable one.
| Layer | Nature | Payback (example) | Reversible? |
|---|---|---|---|
| Technical foundation: crawl, render, structured data | One-time plus upkeep | 2–4 months | Yes, every page benefits |
| Content and answer units | Variable | 3–6 months | Yes, by rewriting |
| Measurement: prompts, monitoring, pipeline | Fixed | None directly | Sets how fast you stop |
| Off-site evidence: reviews, media, communities | Variable | 6–9 months | No |
| People and agency | Fixed | — | Yes, at a cost |
Fund reusable, then reversible, then not. A crawl fix pays across every page and market; off-site evidence is slowest and needs third parties, so it comes after you have something to cite.
That layer carries under-priced risk: a third-party corpus is not ad inventory. You do not control what a reviewer writes, and retractions are not refunded. Budget a sunk amount, not an expected return.
Measurement returns nothing directly, yet without it month three cannot tell you whether to add or stop; see four layers of AI visibility.
Three tiers, not three sizes
A tier is not how much; it is what certainty you buy.
| Tier | Share of recoverable gross profit | What it buys | Move up or stop when |
|---|---|---|---|
| Validate | 5–10% | One testable hypothesis: one market, one category, foundation plus 20 pages | Citation share moves two months running |
| Expand | 15–25% | A repeatable path: off-site evidence, a second market, comparison pages | Payback in one market holds under six months |
| Structure | 30–40% | An operating capability: pipeline, CI checks, multilingual templates | Not an upgrade — another problem |
Percentages are illustrative. Pick gates someone else can recheck — citation share beats a colleague’s sense that AI “started mentioning us.”
A ceiling below one salary points to no dedicated team this year: a part-time owner, a contractor, a finished foundation. The most common first output of a budget model is a decision not to start. Write the exit condition into the proposal before the money moves.
Where this breaks
Cold start. With no traffic to divert, the model outputs near zero; frame it as market-entry cost, not a verdict on the category.
Low AI penetration. A read share near zero will not carry a team: keep light maintenance, put the money into listings and ads.
Marketplace revenue. The site-traffic model does not apply; budget against ad efficiency and listing conversion.
High returns. Margin must be net of returns, or the ceiling inflates and the first refund season overturns it.
Treating it as a promise. Double any parameter and the conclusion flips; the model ranks and caps; it does not commit to growth.
It sits on the strategy and market line: every tier needs a metric that can prove it wrong.
This month, fill the four inputs with sourced numbers, mark the rest “unverified” with a range, and take that page into the budget review. Most of these meetings are decided before anyone speaks.