Updated
How to budget for answer engine optimisation: what is measurable, what is not yet, and how to avoid paying premium rates for ordinary work.
Setting an AEO budget is harder than budgeting for traditional SEO, for one structural reason: the feedback loop is weaker. When an AI assistant cites your page in an answer, you often see no referral click, no query in Search Console, and no clean attribution. You are spending against an outcome you can only partially observe.
That does not make it unbudgetable. It makes it a category where you should fund what is measurable, cap what is not, and refuse to pay premium rates for work that is ordinary SEO with new terminology.
What AEO actually requires that SEO does not
Most of what makes content citable by answer engines is what already made it rank: accuracy, clear structure, genuine expertise, crawlability. The genuinely additive work is narrower than the market implies.
Extraction-friendly structure. Answer engines lift passages. Content organised as clear question-and-answer, with self-contained paragraphs that survive being quoted without surrounding context, can help readers understand the answer. There is no guaranteed citation uplift from a particular paragraph format. Content requiring three preceding paragraphs of setup does not extract cleanly.
Explicit, checkable specifics. Assistants prefer sourceable claims. "Response times improved" is unusable; "p95 latency fell from 800ms to 240ms" is quotable. Numbers, dates, versions, and named entities make a passage citable.
Machine-readable context. Use supported structured data that matches visible organisation, product, or article information. Google does not require special AEO schema. Useful on-page FAQs remain worthwhile, but do not sell FAQ markup as an AI ranking shortcut.
Crawl access for AI agents. Check robots.txt explicitly and distinguish search indexing from model training. OpenAI uses OAI-SearchBot for search and GPTBot for training; Google Search uses Googlebot, while Google-Extended controls separate AI uses. Blocking a training agent does not automatically remove a site from search answers. Set an intentional policy for each agent, then verify access in server logs.
Consistency across sources. Models cross-reference. Contradictory pricing, addresses, or service descriptions between your site, directories, and profiles reduce confidence in all of them.
That is the real list. Anything else sold as AEO deserves scrutiny.
A defensible budget structure
Rather than a flat monthly retainer against an unmeasurable outcome, split spend by how well each component can be verified.
Foundation (one-time, 30–40% of first-year budget). Structured data implementation, crawl configuration, information architecture, entity consistency across external sources. This is verifiable work with concrete deliverables — you can inspect the schema, test the robots directives, and confirm the consistency. Pay for it once and confirm it is done.
Content (ongoing, 40–50%). Depth on topics where you have genuine expertise. This is the component with the clearest return, because good content serves search, AI citation, and human readers simultaneously. It is also the component that fails most often, because the volume-oriented approach that produced results in 2019 now actively harms sites.
Measurement (ongoing, 10–20%). Citation monitoring across assistants, referral tracking from AI sources, brand-mention tracking. Imperfect but improving, and without it you are spending blind.
Reserve nothing for "AEO optimisation" as a standalone line. If a proposal contains that line item without specifying which of the above it means, ask what the deliverable is. The answer is often "we will write blog posts," which is the content line already budgeted.
What you can and cannot measure
Measurable now:
Referral traffic from AI assistants — identifiable in analytics by referrer, though undercounted since many interactions produce no click at all.
Presence in AI Overviews for target queries, checkable manually or with tooling.
Structured data validity, via Google's Rich Results Test.
Crawl access for named AI agents, via server logs — the most direct signal available and the most consistently ignored.
Partially measurable:
Citation frequency across assistants. Tools exist; coverage is inconsistent and results vary by prompt phrasing, session, and region.
Brand mention volume within AI responses, requiring repeated sampling to be meaningful.
Not reliably measurable:
Conversions influenced by an AI answer that produced no click. Someone reads a cited answer, forms an impression, and arrives via direct search a week later. Real, and currently unattributable.
Budget accordingly. Fund the measurable heavily. Fund the unmeasurable at a level you can justify without proof, and be explicit internally that it is a bet.
Rate expectations
Illustrative planning assumptions in USD, not verified market rates or a Voquarn quote. Replace every figure with proposals for your scope:
Technical foundation (one-time) $3,000-12,000
Content, per substantial piece $400-1,500
Ongoing advisory retainer $1,500-6,000/mo
Citation monitoring tooling $100-500/moThe variance is large because the market is immature and the label is unregulated. Two signals of a proposal worth taking seriously: it distinguishes clearly between one-time technical work and ongoing content, and it states what it cannot measure. Proposals promising "AI citation rankings" as a tracked metric are describing something the ecosystem does not currently expose.
Where AEO budgets get wasted
Volume content. Publishing many near-identical pages primarily to manipulate rankings may violate Google's scaled content abuse policy. Page count and use of AI alone do not establish a violation or prove that a traffic drop is a penalty. Evaluate whether each proposed article answers a distinct need with useful evidence; investigate indexing and Search Console reports before assigning a cause to lost visibility.
Schema without substance. Structured data describes content; it does not improve it. Marking up a thin page tells the model precisely what it is.
Chasing every assistant. Coverage differs across systems, and optimising separately for each is not currently a tractable exercise. Write well-structured, accurate, specific content and it performs across systems.
Paying premium rates for standard SEO. Much of what is sold as AEO is competent technical SEO and content work. That work is worth paying for — at technical-SEO rates, not at a premium for the acronym.
A first-year allocation
For a hypothetical B2B site, this example shows how separate line items add up. It does not establish what your business should spend:
Months 1-2 Technical foundation, crawl config,
entity consistency $6,000 one-time
Months 1-12 8-12 substantial pieces $8,000-14,000
Months 3-12 Measurement tooling $2,000-4,000
Months 3-12 Quarterly review + iteration $6,000-10,000
-----------------
~$22,000-34,000The concentration in months one and two is deliberate. Foundation work compounds: everything published afterwards inherits it. Publishing content before the technical base is correct means republishing later.
AEO and GEO deliverables to buy in 2026
Ask the supplier to map each deliverable to an inspectable result: a corrected canonical, a supported schema block, a revised service explanation, or an agreed citation-sampling log. Google’s current generative-search guide says established SEO practices apply and does not require an llms.txt file or special AI markup.
For Pakistan-based suppliers serving international clients, the contract should name the audience, publication language, billing currency, content reviewer, and lead qualification criteria. A mention in an assistant answer is an observation; a completed project enquiry is a business outcome. Report both with their limitations.
Define which pages and topics the engagement covers.
Preserve a baseline of enquiries and source attribution.
Record the region and date of each citation check.
Require correction of inaccurate existing articles before expansion.
Frequently asked questions
How is AEO different from SEO? Substantially overlapping. The additive parts are extraction-friendly structure, explicit checkable specifics, machine-readable context, and deliberate crawl configuration for AI agents. If a proposal claims AEO is entirely separate, it is selling the acronym.
What should we spend? Set a scope-based budget from actual quotes. The example above is arithmetic for one hypothetical allocation, not a market benchmark. Define deliverables and measurement before committing spend.
Can AI citations be tracked? Partially. Referral traffic and AI Overview presence are observable. Citation frequency across assistants is sampleable but inconsistent. Influence without a click is currently unattributable — plan around that rather than buying tools that claim otherwise.
Does publishing more content help? More pages help readers only when they add useful, distinct coverage. Generating pages primarily to manipulate rankings may violate Google's spam policies, but quantity alone does not diagnose abuse or a ranking decline. Prioritise accurate answers, original evidence and clear discovery; neither a large nor a small publication programme guarantees rankings.
Should we block AI crawlers? Decide separately for search crawlers, training crawlers, and user-requested fetchers. Their purposes differ, and allowing access never guarantees a citation. Verify robots.txt reflects your policy and test the relevant user agents.
Further reading
Related
Written by
Moueen Togarvi
Founder & CEO at Voquarn Code, focused on product engineering, search growth, and practical AI systems.