GEA budgets do not look like search budgets, because much of the spend is not media. A realistic split has four lines: evidence work (feed, documentation, proof assets), sponsored placements where the engine offers them, monitoring and tooling, and the strategist time to run the loop. Presenting that split honestly at the start prevents the mid-quarter argument about why so little went to media.
For a first quarter with a new client, a workable starting shape is roughly half to evidence, a quarter to placements, and the remainder split between monitoring and strategist time. As the evidence layer matures the balance shifts toward placements and monitoring, but it never becomes a pure media budget, because evidence decays.
Across engines, allocate by where your clusters actually live rather than by engine popularity. A homeware retailer belongs disproportionately in Amazon Rufus; a B2B software client belongs in ChatGPT and Perplexity. Spreading evenly across engines to look thorough is the most common way to waste a first quarter.
Within an engine, weight toward clusters where you are hedged rather than absent. A hedged mention means you are already in the candidate set and one evidence gap away from a clean citation; an absence may require months of foundational work. Hedged clusters are the cheapest conversions in the plan.
Hold back a contingency, ten to fifteen percent, for displacement response. Competitors change their evidence, engines change their behaviour, and the ability to fund a fast fix in the week it happens is worth more than the incremental placements that budget would have bought.
Report allocation and outcome side by side every month. Showing that sixty percent of the spend went to evidence and that evidence-led clusters carried the citation-share gain is how the evidence line survives the next budget review. Unattributed evidence spend is always the first thing cut.