23SEOGEO · Weekly focus ·
How to evaluate an AI SEO agency: answers, changes, evidence
Evaluate an AI SEO agency using fixed prompts, original answers, citations, page changes and real referrals rather than screenshots or readiness scores.
the 23SEOGEO team · Updated
The short answer
Evaluate an AI SEO agency by asking for original answers and sources for one website and a fixed prompt set, then the exact page to change, implementation owner and review method. Assess mentions, citations, visits, enquiries and sales separately. Compare prices against deliverables, not one flattering screenshot. An internal team can own implementation where capacity exists; outsourcing should fill an explicit gap.
From question to reviewable evidence
Check claims
Choose one website and request reviewable original answers.
- Fix buyer prompts and engines
- Label date, web state and missing evidence
Delivery evidence
Original answers and citation URLs, not selected screenshots.
Open AI Citation TrackerFull action and acceptance checklist
Check claims
Choose one website and request reviewable original answers. Fix buyer prompts and engines · Label date, web state and missing evidence
Original answers and citation URLs, not selected screenshots.
Open AI Citation TrackerDefine delivery
Map each recommendation to a real page and owner. Keep the existing URL and original copy · Use owned facts or verifiable sources
Before/after versions, sources and owner.
Organize owned facts and sourcesReview outcomes
Separate results for the same site and window. Separate mentions, citations and visits · Review enquiries; confirm sales separately
Raw records, visit window and actual publication date.
Review AI traffic and conversions
Start with the problem, not the AI label
Write the buying brief as a specific problem: which customers ask which service questions, in which languages, which page needs work, and whether the business values enquiries or demo bookings. Ask the provider to separate observations, hypotheses and unread data. A visibility chart without fixed prompts, original answers and sample dates cannot be reviewed. Restricted reads are not zero exposure. Google AI features build on SEO foundations; an AI label does not demonstrate a special optimization technique.
Make the monthly report traceable
A reviewable monthly report carries the site, prompt, engine, web-search state, sample date, original answer and citation URLs. Page changes retain the existing URL, before/after copy, implementer and actual publication date. Only completed, non-simulated and verifiable web observations enter measured summaries. Plain model answers and simulations belong separately. English brand aliases may count as mentions, but not as linked citations. GAS explains readiness gaps rather than enquiries, sales or market share.
Compare scope, not an invented market average
Separate research, implementation, content, monitoring, subscriptions and reporting, and identify who owns accounts, raw data and editable files. A report-only engagement is different from implementation and maintenance. Setup and recurring monitoring are different expenses. The uploaded material supplies no original posts or contract evidence for its monetary anecdotes, so this page does not present them as real cases, industry prices or this platform’s offer. Use a current written quote and defined scope.
In-house SEO vs agency: decide by implementation capacity
Internal teams know products, customer questions and business facts, making them suited to owning source material, sales feedback and approvals. An agency can fill audit, editing or implementation gaps. Both should work against one target page and evidence list, without the supplier monopolizing account data. Define a one-page trial, retain the original, and review with comparable market, device and time windows after publication. D+14 and D+28 are suggested observation checkpoints, not promised results.
What evidence supports acceptance?
| Provider material | What it shows | What it cannot replace |
|---|---|---|
| Answer + date + citations | Mentions and citations in a sample | Long-term trends, visits or sales |
| GAS diagnostic | Observable readiness gaps | Measured citation rate |
| Before/after page versions | What actually changed | Proof that rankings rose because of it |
| Site referrals + business review | Visits, enquiries and confirmed sales | Unverified AI attribution |
Related questions
- How to evaluate an AI SEO agency?
- In house SEO vs agency for B2B?
- What should an AI search optimization quote include?
Common questions
Can an agency guarantee ChatGPT recommendations?
No. Prompts, engines, timing and retrieval conditions change. Assess implemented work and reviewable observations rather than a guaranteed recommendation.
Does a report alone mean optimization is complete?
It may complete a diagnostic deliverable. Implementation needs its own record and acceptance.
Is in-house always cheaper?
Not necessarily. Compare actual staff time, capabilities, tools and ongoing maintenance rather than invented costs.
Sources
- Google: AI features and your website
SEO foundations and AI-feature eligibility; no citation or lead guarantee.
- Google: Helpful, reliable, people-first content
For reviewing whether a page serves readers, not an endorsement of a platform score.