An AI search budget that survives Q2 starts with what you can actually measure, not with what a benchmark report promises. There is no trustworthy public dataset yet that tells you what to spend on AEO or GEO in 2026, so the safe move is to size the work against scope and evidence — and to refuse line items that would not survive a calm read-through.
What belongs in an AI search budget (and what does not)
The budget is the sum of work that has a defensible link to a content decision a buyer will eventually take. Eight categories carry real weight; the rest are noise.
Worth funding:
- Technical SEO and crawlability — the floor; nothing else compounds if Google cannot fetch and render the pages.
- Structured data — eligibility work for rich results, framed honestly per the Search Central structured data documentation; not a guarantee of AI citation.
- Content refresh and answer-oriented pages — fewer, sharper pages on the questions your buyer actually asks.
- Topical authority and entity clarity — internal linking, naming, and the things your About page makes obvious.
- Analytics and Search Console workflows — including the new Search Console AI views where available (limited rollout).
- AI visibility monitoring — only when you can name what decision it will inform.
- Product or feed data — for ecommerce, the catalogue quality budget is the AI search budget for a large share of queries.
- Editorial QA and internal workflow — the unglamorous middle that keeps the rest honest.
Worth refusing:
- Generic blog volume with no evidence plan attached.
- Random “AI tool” subscriptions adopted before a measurement question exists.
- “LLM submission” services that promise inclusion in model training.
- Citation-guarantee services for AI Overviews and similar surfaces.
- Vanity dashboards that do not drive a decision anyone has named.
A conservative planning model
Six moving parts. No dollar figures pretending to be benchmarks; specific numbers belong in your own forecast, anchored to your own audit.
- Baseline audit. Inventory what you have, what ranks, and what already shows up in AI surfaces for branded and high-intent queries.
- Measurement setup. Make sure Search Console + GA4 + (if appropriate) a GA4 raw events export to BigQuery are healthy before the spend begins.
- Priority content refresh. A short list of pages where a refresh changes the decision a real buyer takes, not just the search query that triggered them.
- Entity and schema fixes. Author, organisation, product entities present and consistent across the pages that matter.
- Experimentation queue. Two or three small bets per quarter, each with an explicit hypothesis and an explicit failure condition.
- Monthly review loop. What changed; what we logged; what the dashboard says vs what the off-platform data says.
This is a planning model. Anyone who hands you the same template with confident percentages attached has invented the percentages.
Reading a budget by area (table)
The table below is the conversation I have with a client before signing a scope. The right-hand column is the line item I will not let through.
| Budget area | What it funds | When it is necessary | What to avoid | Evidence needed before scaling |
|---|---|---|---|---|
| Technical SEO | Crawl, render, performance, internationalisation | Always — it is the floor | Treating it as a one-time project | Stable index coverage and Core Web Vitals trend before adding spend |
| Structured data | Eligibility for rich results; Google-parsable entities | When you have content that maps cleanly to documented types | “Schema guarantees AI citations” — it does not | Validated markup + a control group of unmarked pages |
| Content refresh | Specificity, evidence, and answer clarity on pages that matter | Where the page is already in the consideration set | Volume targets without a refresh logic | Reading-time, internal-link, and conversion movement on the refreshed set |
| AI visibility monitoring | Spot-checks of AI Overview / Discover AI presence | Only when you can name the decision the data will trigger | Buying a “score” with no methodology page | A written decision rule before subscribing |
| Measurement architecture | Search Console, GA4, BigQuery, dashboards | Before anything downstream is scaled | Dashboards with no owner | Reconciled numbers across at least two surfaces |
| Experimentation | 2–3 small bets per quarter, each with a written hypothesis | Once measurement is honest | “Always be testing” with no design | Closed-loop write-up per experiment, success or otherwise |
How budget changes by company type (qualitative, not banded)
Different operating realities, different shapes. The numbers are intentionally absent — anyone giving you universal bands is selling them.
- Local service business. Disproportionate value from technical hygiene, GBP/listing data, and a small set of answer pages tied to a calendar of services. Most “AI visibility tool” spend is premature.
- B2B SaaS. The budget concentrates on category clarity, entity-level work for products and people, and a documentation hub that an AI surface can quote without paraphrasing badly.
- Ecommerce. Product feed quality and structured data carry more weight than blog volume; AI surfaces lean on the catalogue for transactional queries.
- Enterprise / multi-market. The dominant cost is coordination — the structured data and content systems across markets, not the headline AI investment.
- Agency portfolio. The portable cost is the playbook and the measurement template, not any single client’s tool subscription.
Measurement before scaling
Two practical baselines before the budget grows.
- Search Console baseline. Export query and page clusters at the same time window; include the new AI views if your property has them. The new Search Console AI report rollout is partial, so do not assume your account is in it.
- Off-platform truth surface. GA4 plus, where useful, the raw events in BigQuery. This is where the click actually becomes a decision; the ad/search platforms are upstream.
For the academic context that keeps people honest, the 2026 Xu et al. preprint covered 55,393 trending queries and decomposed responses into 98,020 atomic claims; AI Overview activation was 13.7% in that sample, and 11.0% of atomic claims were unsupported by their cited pages. Those are study-scope findings, not universal benchmarks. Google’s own AI features and your website documentation remains the reference for which surfaces use your content.
Red flags in GEO/AEO budget proposals
If a proposal contains any of these, it is a sourcing problem, not a budget problem:
- Guaranteed AI Overview citations.
- A proprietary “AI visibility score” with no methodology page.
- Benchmark numbers with no sample details, time window, or methodology.
- Content volume promises with no measurement clause.
- No distinction between SEO, GEO, CRO, and paid media — the line items collapse into one bill.
- Single-number ROI claims for AI search.
Open questions
Where the answer is honestly “we do not know yet”:
- Public AI search measurement is limited; nobody has a universal denominator.
- AI Overview behaviour shifts; a benchmark from January is not a benchmark in June.
- There is no shared industry standard for what counts as “AI search visibility” — every vendor defines it differently.
- Attribution between AI visibility and revenue is unresolved at any reasonable scale.
- Vertical and market differences are large; a B2B SaaS in Berlin and a clinic in İzmir share little.
If you would like a second pair of eyes on the budget brief — the scope, the measurement plumbing, the experimentation logic — that is what my marketing audit and consulting engagement covers. Longer essays on AI search measurement and GEO sit in Insights, and the GEO Audit Skill in the library is the structured starting point I use during an evidence-first review.
Frequently Asked Questions
Is there a reliable industry benchmark for AI search budgets in 2026?
Not yet. Publicly cited “AEO/GEO benchmark” datasets are mostly agency surveys with limited sample disclosure. None of them is in our Fact_Bank, and none is a substitute for an audit of your own scope and measurement.
What is the smallest credible AI search budget?
The smallest credible budget is whatever covers technical hygiene, one round of priority content refresh, and a measurement loop that can detect change. The dollar figure depends entirely on your scope; specific bands without scope are noise.
Does structured data guarantee AI Overview citations?
No. Google’s own structured data documentation positions schema as eligibility work for rich results, not a guarantee of AI surface inclusion. Treat it as a discipline that makes your content easier to parse — not a citation contract.
Should I buy an AI visibility monitoring tool?
Only when you can name in one sentence the decision the data will trigger. If the tool ships before the decision exists, it ships as a vanity dashboard.
How do I size the experimentation line in an AI search budget?
Two to three small bets per quarter, each with a written hypothesis, a written failure condition, and an agreed write-up after the cycle. The line item is the discipline more than the dollar amount.