How we test generative engine optimization tools
The buyer types, criteria, weights and evidence rules behind every Best GEO Platform comparison, published before any ranking.
Published
We score each GEO platform on eight criteria, from 0 to 4, and weight the scores for one of four buyer types. Every score says how we know it: tested, researched or vendor claim.
Only a tested result can score 4, and we rank a product only when tested criteria carry at least 60 percent of the weight. Every platform gets the same questions, brands and dates, checked against our own baseline. We published these rules before any comparison.
Four buyer types
Each comparison is written for one of these buyers:
- Small businesses: one brand, a small budget, no specialist on staff.
- In-house marketing teams: one company, a few people, reports for management.
- Agencies: several client brands and client-facing reports.
- Enterprises: many brands or markets, security reviews and procurement.
Eight criteria
We score each criterion from 0 (missing or failed) to 4 (the best we have seen).
| Criterion | What we check |
|---|---|
| Engine coverage | Which AI engines and surfaces does it track, and did our test confirm them? |
| Measurement design | Does it separate mentions from citations, show the cited sources, repeat prompts, and let you set location and language? |
| Agreement with our baseline | Do its results match our own runs of the same questions on the same days, through routes the AI services allow? |
| Actionability | Does it tell you what to change, and is that advice specific to your pages and your prompts? |
| Reporting and workflow | Exports, scheduled reports, several brands or clients, white-label reports. |
| Setup and ease of use | How long from sign-up to the first useful data, and how much expertise does it need? |
| Price and terms | Cost for a defined workload, trial terms, minimum contract. |
| Data access and governance | API, single sign-on, user roles, data retention and export. |
Each buyer type weighs the criteria differently. Price and setup count most for a small business, client reporting for an agency, and access control for an enterprise.
| Criterion | Small business | In-house team | Agency | Enterprise |
|---|---|---|---|---|
| Engine coverage | 15 | 15 | 15 | 15 |
| Measurement design | 15 | 20 | 15 | 20 |
| Agreement with our baseline | 15 | 15 | 15 | 15 |
| Actionability | 15 | 15 | 10 | 10 |
| Reporting and workflow | 5 | 10 | 25 | 10 |
| Setup and ease of use | 15 | 10 | 5 | 5 |
| Price and terms | 20 | 10 | 10 | 5 |
| Data access and governance | 0 | 5 | 5 | 20 |
Tested, researched or unknown
- Tested We used the product, ran the task and logged the output with the date.
- Researched From the vendor’s documentation, pricing page or another primary source. Not run by us.
- Vendor claim The vendor says so and we couldn’t check it.
- Unknown We don’t know, so we don’t guess.
A researched score can reach 3 at most and a vendor claim 1. We rank a product for a buyer type only when its tested criteria carry at least 60 percent of that buyer’s weight. Below that, it is listed as “not yet tested”.
How a test runs
- We choose a product category, the brands to track and a set of buyer questions, and publish all three.
- Every platform gets the same questions, brands, country and language on the same day, on the plan we name.
- We compare each platform’s results with our baseline: the same questions, run on the same days.
- Each platform profile keeps a test log: the task, the output, the date, the plan, the price and the gaps.
The baseline
We run the baseline only through routes the AI services’ terms allow: their official APIs, labeled “API”, or a person reading the results, labeled “manual”. API answers can differ from what the consumer apps show. So we compare patterns over several days, such as which brands and sources come up, rather than single answers.
Prices and features
Prices come from the vendor’s public pricing page, with the date, the plan and the billing basis. We check prices and features again before we publish and at least every 90 days. When something changes, the page says what and when.
Money and mistakes
No vendor can pay for a place in a ranking or a better score. Any affiliate link, free access or other relationship is disclosed next to the recommendation it affects; see our disclosure policy. If we get something wrong, we fix it and note the change on the page. Report errors to hello@bestgeoplatform.com; see our corrections policy.
Changes to this method
- : First draft. Published before any comparison or ranking.