Short answer, as of August 2026: hire on methodology, not credentials, because nobody has a long track record in a discipline this young. Expect to pay roughly $1,500 to $6,000 a month for a genuine engagement, or $99 to $999 a month if what you actually need is tracking software rather than a person. The one question that filters most of the market: "Which deliverables in this proposal would a competent SEO program not already produce?" If the honest answer is "measurement," you need a tool.
Here is how to run the rest of the evaluation.
Why the usual hiring signals don't work here
Generative engine optimization became a commercial category around 2023, when the Princeton-led GEO paper gave it a name and a benchmark. That means:
- Nobody has ten years of experience. Anyone claiming otherwise is describing SEO experience, which is relevant but is not the thing they said.
- Case studies are thin and hard to verify, because the outcome — being named in a non-deterministic answer — leaves no public artifact a prospect can independently check.
- Certifications are meaningless. No standards body accredits this. Any badge on a proposal was self-issued or bought.
Strip those away and what remains is method. Can they explain how they will measure, what they will change, and why they believe the change causes the outcome? That is the whole evaluation.
The seven qualifying questions
Ask these verbatim. The quality of the answers separates operators from repackagers faster than any portfolio review.
1. "How will you measure whether this worked?" You want: a fixed prompt set, sampled repeatedly across multiple engines, tracked over time as share of voice. You do not want: "we'll monitor your rankings" (wrong surface) or "you'll see it in your traffic" (structurally unmeasurable — see below).
2. "How many times will you sample each prompt, and across which engines?" AI answers are non-deterministic. A vendor who checks each prompt once has built a random number generator. The right answer names repetition counts and specific engines — ChatGPT, Gemini, Perplexity, Claude, Grok, Microsoft Copilot — because coverage differs sharply between them.
3. "How will you attribute results, given citations often produce no referral traffic?" The correct answer acknowledges the problem: an assistant names you, the user calls you or searches your brand, and analytics record a direct visit. A vendor who promises clean last-click ROI on AI citations either hasn't done the work or is willing to mislead you about it. Honest instruments are share of voice, branded search volume in Search Console, and self-reported attribution at intake.
4. "What will you check before writing any content?" You want to hear the mechanical audit: AI crawler access in robots.txt and at the WAF, server-side rendering of critical content, entity consistency across profiles and directories. A vendor who leads with a content calendar is skipping the diagnosis — and content is worthless if the crawler can't fetch the page. The full sequence is in our diagnostic for when a competitor is cited and you aren't.
5. "What does your content standard actually require?" Good answers cite the research: specific sourced claims, statistics, direct quotations, named exceptions — the modifications the Princeton benchmark found lifted generative-engine visibility by up to 40%, versus keyword stuffing, which did nothing. Bad answer: "we write high-quality, engaging content optimized for AI." That sentence has no operational content.
6. "Who writes it, and how do you get facts only my business knows?" The specifics that make a page quotable — actual durations, actual exceptions, actual constraints — live inside your operation. A vendor with a real process has an extraction mechanism: subject-matter interviews, a structured intake, a review loop with your technical staff. A vendor without one will produce competent, generic, uncitable prose.
7. "What happens to this work if I stop paying you?" Content, structured data and a fixed entity record are assets you keep. A monitoring dashboard is a subscription you lose. Both are legitimate; you should just know which you're buying.
What should be in scope
| Deliverable | In a real GEO engagement? | Notes |
|---|---|---|
| Prompt-set design and baseline | Yes | The foundation — without a baseline nothing is attributable |
| Recurring multi-engine sampling | Yes | Usually delivered via software, not manual checks |
| AI crawler access audit | Yes | robots.txt and the WAF/CDN layer |
| Rendering audit | Yes | Raw HTML response, not the browser DOM |
| Entity consistency audit + cleanup | Yes | Slow, unglamorous, highest-leverage |
| Content upgraded to the citation standard | Yes | Prioritise pages that already rank |
| Structured data generated from one record | Yes | Hand-authored per-page markup drifts |
| Digital PR / third-party corroboration | Sometimes | Slow and expensive; scope it explicitly |
| "Guaranteed placement in ChatGPT" | No | Not a product the engines sell |
| "AI-optimized keyword density" | No | Directly contradicted by the research |
If a proposal is mostly rows from the bottom of that table, you are looking at repackaged SEO with a premium attached.
Five red flags
Guaranteed citations. Organic inclusion in an AI answer is not purchasable. Paid placements in AI surfaces exist, are labelled as advertising, and are bought from the platforms — not from a consultant.
A single before/after screenshot as proof. One prompt run twice proves nothing given non-determinism. Ask for a time series across repeated samples, or treat the evidence as anecdote.
No mechanical audit in the proposal. If crawler access and rendering aren't in scope before content, the vendor is skipping the two causes that make everything else pointless.
Volume-priced content. "Twenty AI-optimized articles a month" is a content mill. Generative engines reward depth and specificity on pages that get retrieved; twenty shallow pages is a worse outcome than three excellent ones, and it's a worse outcome than doing nothing.
Vagueness about engines. A specialist should be able to tell you offhand that engine coverage differs, which ones lean on which indexes, and why a client might appear in one and not another. If every engine is "AI" to them, they haven't looked closely. We covered those differences in how the major engines pick their sources.
Specialist, generalist agency, or software?
Three different situations, and the wrong choice is expensive.
Buy software when your SEO fundamentals are healthy, you have people who can write, and what you actually lack is visibility into whether AI engines name you. This is the most common real situation and the most commonly over-solved. Tracking runs roughly $99 to $999/mo depending on how many brands or client tenants you cover.
Hire a specialist when you've run the diagnostic, found a genuine content or corroboration deficit, and need someone to execute at a standard your team can't currently hit. Expect $1,500 to $6,000/mo, and expect the first month to be audit rather than output.
Extend your existing SEO agency when they can credibly answer the seven questions above. Many can — the work is largely adjacent to what they already do, and a vendor who already knows your business beats a new relationship that has to learn it. Ask the questions anyway; some will fail them, and that is useful to know.
The comparison we wrote between platform, BrightLocal and a traditional agency works through the same trade-off for local SEO, and the economics rhyme closely.
How to structure the first 90 days
Whoever you hire, insist the engagement is sequenced like this. A vendor who resists this order is optimising for looking busy.
- Weeks 1–2: baseline and audit. Prompt set defined and sampled, crawler access confirmed, rendering verified, entity inconsistencies catalogued. Deliverable: a written diagnosis naming which of the five failure causes you actually have.
- Weeks 3–6: mechanical fixes. Crawler access, rendering, structured data, entity cleanup. Cheap, fast, and prerequisite to everything else.
- Weeks 7–12: content. Upgrade the pages that already rank before writing new ones — retrofitting a retrievable page is far higher-return than earning retrieval from scratch.
- Ongoing: monitoring and off-site. Recurring sampling so regressions surface as a chart, and the slow corroboration work measured in quarters.
If month one is a content calendar, you have hired a writer, not a specialist. That's fine if it's what you need — just don't pay specialist rates for it.
When you mostly need the tooling
For a lot of businesses the honest conclusion is that the fundamentals are fine, the writing is already handled internally, and the gap is purely that nobody knows what the engines are saying. DigiRank's AI Visibility Tracker exists for exactly that case: define a prompt set, sample it across ChatGPT, Gemini, Perplexity, Claude, Grok and Microsoft Copilot on a schedule, and get share of voice per prompt over time — with the multi-tenant white-label reporting agencies need on the $249/mo Agency plan, which includes a 14-day trial. It connects to Search Console, GA4 and Google Business Profile so AI-side visibility sits beside conventional performance data.
If you want the underlying economics before you talk to anyone, what generative engine optimization actually costs itemises the line items, and how to split a budget between GEO and traditional SEO covers the allocation question that usually follows.
Frequently asked questions
How do I find someone who specializes in generative engine optimization? Evaluate on methodology rather than credentials — the field is too young for meaningful track records and no body certifies it. Ask how they measure (a fixed prompt set sampled repeatedly across multiple engines), what they audit before writing content, and which deliverables a competent SEO program wouldn't already produce.
What should generative engine optimization cost? Roughly $1,500 to $6,000 a month for a genuine service engagement including audit, content and off-site work. If your fundamentals are healthy and you only need visibility into what engines say about you, tracking software at $99 to $999 a month is usually the correct purchase instead.
Can a GEO agency guarantee I'll be cited by ChatGPT? No. Organic inclusion in an AI-composed answer is not something the engines sell. Paid placements in AI surfaces exist but are labelled advertising and bought from the platforms directly. A guarantee of organic citation is a reliable signal to walk away.
Is my existing SEO agency able to do this? Often yes — most GEO work overlaps heavily with competent SEO. Ask them the qualifying questions about measurement, engine coverage, the pre-content mechanical audit and their content standard. An agency that already understands your business and answers those well beats starting a new relationship.
What deliverables should be in a GEO proposal? Prompt-set design and baseline, recurring multi-engine sampling, an AI crawler access audit covering both robots.txt and the WAF, a rendering audit of the raw HTML response, entity consistency cleanup, content upgraded to a citation standard, and structured data generated from a single canonical record.
How long before a GEO engagement shows results? Mechanical fixes surface in days to weeks. Content upgrades on already-ranking pages take weeks. Entity consistency propagates slowly because directories cache. Third-party corroboration takes quarters. A vendor promising visible AI citation changes in the first month is overselling.
What's the difference between hiring a specialist and buying software? Software gives you measurement and automation of repetitive work — prompt sampling, audits, scheduled reporting. A specialist gives you judgment and execution: deciding which pages matter, extracting the specifics only your business knows, and doing the slow off-site corroboration work. Most companies with healthy fundamentals need the software first.