Choosing GEO software is now hard to answer with a single sentence about “which one is best.” The “12 platforms covered,” “5 mainstream AI,” and “8 major AI engine endpoints” written on different vendors’ official websites are not the same unit of measurement—line them up and subtract one from another, and the conclusion you get will most likely be inaccurate.
This article is not a review and does not rank anyone. It does one thing only: it copies down, using one identical set of fields, the information that 8 GEO software vendors publish on their own official websites, and labels the actual page URL each item came from, the scope of this crawl, and the verification time (2026-09-04 PDT), so that you can judge for yourself. We also put ourselves into the same table, using the same fields, including the numbers that do not look good.
1. Why GEO Software Has to Be Compared Across the Board First
Traditional SEO has at least one shared reference point: the ranking position on a search results page. AI search has none. Ask the same question and Doubao answers one way, DeepSeek another, and Tencent Yuanbao a third; the answers are generated, not a list that can be aligned position by position.
So what you are really buying when you buy GEO software is a measurement methodology—who gets to define what “my brand was mentioned” means. Different methodologies produce figures that are not comparable. There are three places where it is easiest to step into a hole.
First, “number of engines” is not one single unit. Some vendors count models, some count “engine endpoints” (model × access endpoint), some count “platforms,” and some mix domestic and overseas into one total. Two pieces of original copy from this crawl: 透镜 GEO’s pricing page says “5 mainstream AI: Doubao · DeepSeek · ERNIE Bot · Qwen · Yuanbao” and separately labels “web / APP / both,” while its home page says “covers 8 major mainstream AI engine endpoints” (both seen on geo.timus.cn, verified 2026-09-04 PDT; the official website does not explain how the two convert into each other). Zhituishidai GenOptima carries three numbers on one and the same home page: “full coverage of 20+ AI search platforms worldwide,” “deep adaptation for 14 large models (6 domestic + 8 overseas),” and “covers 6 major AI platforms” (zhituishidai.com, same verification time). When one company writes both 6 and 20 on the same page, the question to settle before any comparison is what each of those two numbers is counting.
Second, the billing unit is not one single unit either. Among the 8 vendors verified here, three completely different commercial forms coexist: subscription by number of monitored questions (透镜 GEO Free ¥0 / Starter ¥79/month / Basic ¥199/month / Professional ¥499/month), one-time billing per report (SOUGEO 9.9 yuan Express / 199 yuan Professional / 399 yuan Dynamic Analysis, not monthly), and billing by service budget and annual framework (Zhituishidai Lite ¥9,800/month, with the SkaaS section separately listing an annual-framework average deal size of “30-200w”). In other words, “¥199” and “199 yuan” may be a monthly fee in one case and a single one-time report in the other. These prices are billed in mutually incommensurable units, and this article only places the original wording side by side; it performs no normalization (it does not annualize report prices, and it does not multiply monthly fees by dispatch volume), because the engine counts and frequency methodologies that any normalization would require are not stated clearly by the vendors themselves.
Third, official websites are not always internally consistent either. In this round of verification, two vendors had three mutually inconsistent coverage numbers within a single page (AIDSO home page: 6 named vendors ／ “10-plus” ／ “12 platforms”; Zhituishidai home page: 20+ ／ 14 ／ 6); another three were inconsistent across pages, or between a page and an interface (透镜 GEO’s home page “8 engine endpoints” versus its pricing page “5 AI”; Dageno’s pricing page offers a selectable list of 7 while its footer lists only 6; ImpetaAI’s Enterprise plan returns modelLimit=5 from the interface while the page copy says “customized on demand”). Our handling is always to place them side by side exactly as written, without adjudicating which one counts—that is something the official websites need to clarify, not a judgment for us to make.
There is one more reason, and it comes from a record of our own. On 2026-08-29, we searched for our own brand inside Doubao, and the answer that session gave stated that zhimahang.com is not the official website of “引见 YinJen” and advised the user to verify in order to avoid entering an imitation site. This needs to be stated plainly: it was a single-session observation, not a reproducible experiment—ask the same question again, at a different time, or from a different account, and the result may differ; we ran no repeated retries and no statistics. What it tells us is only this: writing content on your own site does not mean it can be cross-verified from outside. That being so, we have to hold to the same line even more strictly when writing about others—every item of information must be traceable back through its link.
2. The Evidence Methodology and Evidence Scope of This Article
- Verification time: 2026-09-04 (PDT). Prices and engine lists are live and can change at any moment; this article is accountable only for that moment.
- How evidence was collected: we used a desktop browser UA to curl each vendor’s public pages directly. All returned HTTP 200, and not one depended on a logged-in session. For one of them (ImpetaAI 慧辰), the prices are not written into the HTML—the page shows a placeholder ¥0, and the real figures are served by that site’s own public interface, which requires no login. This article takes them from that interface and notes the path.
- The crawl scope is limited, so this article draws no exhaustive conclusions. Sites of this kind are mostly single-page applications, and body text and prices often live inside a JS bundle or an interface. Every “does not appear” or “not retrieved” in this article is therefore accountable only for the URLs listed in the source list at the end of Section 3; it does not mean “this does not exist anywhere on that site.” Which vendors had their JS bundles / structured data / interfaces searched, and which had only HTML retrieved, is written out line by line in that list.
- Only self-descriptions from official websites are accepted: no third-party reviews, ranking lists, channel-partner marketing material, or community hearsay.
- Three rules we hold ourselves to: no scoring, no ranking, no speculation about motives, no evaluation of who is better or worse. Any attribute for which we cannot produce verifiable evidence is deleted rather than written up as “reportedly.”
- If we could not get it, we say we could not get it: four items were explicitly not obtained this time—Profound’s monthly price (the page has a monthly/annual toggle, and the monthly figure is rendered by script), the names of the 5th and 6th platforms in the logo carousel on the Newrank Geowise product site (the alt attributes are only img1–img6), AIDSO’s GEO product pricing within this crawl scope, and the full legal names of the operating entities behind Profound / SOUGEO / Dageno / Zhituishidai (their pages carry only an ICP filing number or an address). All of these are labeled as such, with no guessing.
There is one more note that applies to all GEO software (ourselves included): “supports N engines” on an official website is usually a configuration methodology—that is, the product has integrated N of them. It does not mean your account will get N sets of valid data back in every round. There are many links in the chain that can break: account login sessions, interface rate limits, changes on the engine side. Configuration count and measured valid samples are two different numbers, and they are worth asking about separately when you look at data. We are ourselves a sample of that difference, and Section 4 writes it out. Also, most GEO software states engine coverage by plan tier, so any citation must carry which tier it refers to—otherwise you are inflating the other party’s number on their behalf.
3. Public Information from 8 GEO Software Vendors, Side by Side
(Row 9 of the table below is us, using the same set of fields. The complete source URL for every claim in every row is in the evidence-scope list at the end of this section; the verification time throughout is 2026-09-04 PDT.)
| Product (domain) | Operating entity (website footer / self-description) | Pricing public? | Engine/platform coverage methodology as written on the official website (copied verbatim, by tier) | Free plan / trial |
|---|
| AIDSO 爱搜 (aidso.com) | Hangzhou Aisou Quanyu Data Technology Co., Ltd. (杭州爱搜全域数据科技有限公司; from the footer copyright line; the footer address is Chaoyang District, Beijing) | No product pricing retrieved within this evidence scope | Three figures side by side on the same page: 6 named vendors (Doubao / DeepSeek / ERNIE Bot / Qwen / Kimi / Tencent Yuanbao) ／ “10-plus mainstream AI platforms” ／ “covers 12 platforms”; no complete named list | The words “free” and “trial” do not appear in the 2 pages crawled |
| 透镜 GEO (geo.timus.cn) | Beijing Shiguang Buyu Technology Co., Ltd. (北京时光不语科技有限公司); ICP filing 京ICP备2025155669号-1 | Public, with a standalone pricing page | Pricing page, paid tiers: “5 mainstream AI: Doubao · DeepSeek · ERNIE Bot · Qwen · Yuanbao” + “web / APP / both”; the home page separately says “8 major mainstream AI engine endpoints”; both methodologies placed side by side | Free plan ¥0 (3 monitored questions, 3 web-side engines) |
| Zhituishidai GenOptima (zhituishidai.com) | The footer address is Jing’an District, Shanghai (no full company name seen on the page) | Partly public: the Lite tier is priced, the other two tiers are customized; no standalone pricing page (/pricing returns 404) | Three figures side by side on the same page: “full coverage of 20+ AI search platforms worldwide” (listing 8 overseas + 12 domestic item by item) ／ “14 large models (6 domestic + 8 overseas)” ／ “covers 6 major AI platforms” | The words “free” and “trial” do not appear in the home-page HTML crawled |
| Newrank Geowise (geo.newrank.cn) | Part of Newrank (newrank.cn) | No product price seen within this crawl scope; every CTA is “request a demo” | Newrank’s main site says verbatim “already supports 6 major AI platforms including Doubao / Yuanbao / DeepSeek / Kimi” (4 named, total stated as 6); the product site’s first screen is a carousel of 6 text-free logos (alt attributes img1–img6), and the names of the 5th and 6th could not be retrieved this time | One news card on the product site reads “Request a demo appointment and receive 7 days of trial access!” |
| Profound (tryprofound.com) | No operating-entity information retrieved from the official website this time | Public, with a standalone pricing page | By tier: Starter “ChatGPT tracking only” (1); Growth “3 Answer Engines tracked” = ChatGPT / Perplexity / Google AI Overviews; Enterprise “Up to 9 Answer Engines tracked,” with 9 named in the comparison table | The Growth tier’s CTA is “Try for free”; the official website states no number of days |
| ImpetaAI 慧辰 (impetaai.hcr.com.cn) | Beijing Huichen Zidao Information Co., Ltd. (北京慧辰资道资讯股份有限公司) | Public, but the prices are served by the official website’s own public interface | The interface gives a named list per tier: Free 3 (Doubao / DeepSeek / ERNIE Bot); Advanced and Professional 6 (＋Kimi / Yuanbao / Qwen); for Enterprise the interface returns modelLimit=5 with 5 named, while the page copy says “customized on demand”—the two are placed side by side as written | Free plan ¥0; the interface returns freeDays=7 |
| SOUGEO 搜极星 (sougeo.com) | Footer ICP filing 粤ICP备2026003264号-1 (no full company name seen on the page) | Public, but billed by report tier | FAQ text: “covers 12+ mainstream large-model platforms across the web,” with 7 domestic ones named (DeepSeek / ERNIE Bot / Qwen / Kimi / Doubao / Zhipu Qingyan / Yuanbao); overseas ones not named item by item | The words “free” and “trial” appear neither in the home-page HTML crawled nor in its FAQ source file |
| Dageno (dageno.ai) | English-language site; no operating-entity information retrieved from the official website this time | Public, with a standalone pricing page | All three paid tiers are labeled “Platforms: 3,” and the comparison table says “Choose any 3 platforms,” from a selectable pool of 7 (ChatGPT / Grok / Gemini / Perplexity / Google AI Mode / Google AI Overview / Copilot); Enterprise is “Custom full coverage”; the footer lists only 6 items (no Copilot), placed side by side with the pricing page | The top of the pricing page says verbatim “All plans include a 7-day free trial” |
| YinJen 引见 (zhimahang.com/yinjen) | Suzhou ZhiMaHang Technology Co., Ltd. | Public | External methodology: 12 AI engines (a configuration count, i.e. 12 have been integrated); in our own self-monitoring, 11 are measured as valid—claude has produced nothing for 5 consecutive rounds because the login session on our account expired, which is a problem with our account and unrelated to that engine’s service | 14-day free trial |
Vendor-by-vendor notes (all are official-website self-descriptions; the verification time throughout is 2026-09-04 PDT)
AIDSO 爱搜 (source: https://www.aidso.com/ ): its meta description describes it as “a leading domestic comprehensive service provider for DSO full-domain search optimization and GEO optimization,” claims a “DSO+GEO dual-engine” strategy, covers content platforms such as Douyin and Xiaohongshu along with AI platforms such as DeepSeek and Doubao, and offers data monitoring, training, and managed operation services. On pricing, the evidence scope has to be stated precisely: this time we retrieved only 2 HTML responses, the home page and https://www.aidso.com/pricing , and did not search the site’s JS bundle or structured data; /pricing returned 200, but its response body was exactly the same byte count as the home page (341552 B), from which we judge it to be a single-page application’s fallback page, so no standalone pricing content was retrieved at that path this time. In the HTML of those two pages, the only price figure that appears is in the footer: “Knowledge Planet, 365 yuan/year, our instructor team answers your questions all year round”—that is a knowledge-community subscription, not a price quote for the GEO software product, and it must not be cited as a product price. Our conclusion about this vendor can therefore only be: no GEO product pricing was retrieved within this evidence scope, which does not mean the site has no public pricing.
透镜 GEO (source: https://geo.timus.cn/pricing and the home page): the prices are printed directly in the page HTML—Free ¥0, Starter ¥79/month, Basic ¥199/month, Professional ¥499/month, and Custom priced on demand. Term discounts (FAQ text): “3 months at 0.8× the price, 12 months at 0.7× the price, applicable to all paid tiers.” Tiers are divided mainly by number of monitored questions: Free 3, Starter 8, Basic 30, Professional 100. The free plan’s boundaries are written out in some detail: 3 monitored questions and 3 web-side AI engines, including real-account collection ／ brand ranking analysis ／ brand sentiment analysis, and excluding search-snapshot downloads, citation-source intelligence insights, and data-export reports. Monitoring is subscribed per brand, while copywriting and publishing are billed separately by credits; the official website says credits are deducted on successful publication and refunded in full on failure. The site also appends a disclaimer: “The above are the AI models currently supported; coverage may be adjusted dynamically in line with each platform’s situation, and the actual service prevails.”
Zhituishidai GenOptima (source: https://zhituishidai.com/ ): the first screen describes it as “the brand consensus operating system for the AI search era,” with a SkaaS (Skill as a Service) model, stating that “143 GEO expert capabilities are invoked on demand as cloud-based skills.” The three tiers, verbatim: 智推轻选 Lite at “¥9,800 / month” (the body text says “standardized GEO service, monthly budget starting at 10,000 yuan”); the standard GenOptima tier is “a customized plan ／ pricing quoted on demand according to industry and growth targets”; and the Ultra flagship tier is “dedicated customization,” aimed at enterprises with revenue above 10 billion yuan; the SkaaS section separately lists an annual-framework average deal size of “30-200w.” The billing unit at this tier—a monthly service budget and an annual framework—does not belong to the same category of procurement as tools subscribed by question count, and the two cannot be read side by side.
Newrank Geowise (source: https://geo.newrank.cn/ ; the engine-count methodology is taken from the group’s main site https://newrank.cn/ ): first, the relationship between the two domains—geo.newrank.cn is the product site and newrank.cn is the group’s main site, whose GEO section entry points at the product site; this article treats the two as separate evidence objects and labels them separately. The product site’s first screen reads: “organizational enablement of cross-team GEO capability for medium and large brands,” and its feature modules include prompt mining, scenario lookup, continuous tracking and iteration, creative optimization, and reverse lookup of citation rates. No product price of any kind was seen in the product-site HTML crawled this time (the “0 yuan” that appears on the page is the title of an on-site news card, “Run the Full GEO Chain for 0 Yuan: You Can Do It Yourself Without an Agency,” not a product price). One more line that can be checked: the value-added services listed in the main site’s GEO section are “systematic training, field study visits, digital large screens, consulting reports, and custom APIs,” likewise unpriced—this only says “the official website lists these value-added services,” and this article draws no inference from it about where the business model’s center of gravity lies.
Profound (source: https://www.tryprofound.com/pricing ): an overseas product, included here for comparison. The pricing page has two tabs, For brands and For agencies; Starter is $99/month and Growth is $399/month, both labeled “Billed yearly” and “2 months free,” with Enterprise as Custom. The engine counts are the official website’s own wording printed in its tier comparison table, not figures this article calculated: the Starter column says “ChatGPT tracking only,” Growth says “3 Answer Engines tracked” (the comparison table names ChatGPT, Perplexity, and Google AI Overviews), and Enterprise says “Up to 9 Answer Engines tracked” (the Capability for column names 9: ChatGPT, Perplexity, Google AI Mode, Google Gemini, Microsoft Copilot, Grok, DeepSeek, Anthropic Claude, and Google AI Overviews). Collection volumes are stated on the official website: Starter 50 unique prompts / 1,500 responses monthly; Growth 100 prompts / 9,000 responses monthly; frequency Daily; the Growth tier has 1 language and 1 region. It should be noted that the page has a monthly/annual toggle; what this static crawl retrieved is the annual methodology, and the monthly price is rendered by script and was not retrieved.
ImpetaAI 慧辰 (source: https://impetaai.hcr.com.cn/marketing/intro ; the prices and engine list are taken from the same site’s public interface https://impetaai.hcr.com.cn/api/plan/showList ): launched by Beijing Huichen Zidao Information Co., Ltd. (北京慧辰资道资讯股份有限公司), whose “About Us” page states it was founded in 2008; its six modules are brand-awareness monitoring, competitor recommendation-ranking analysis, prompt question-matrix analysis, citation-source identification, sentiment and risk analysis, and intelligent diagnostic reports. How the prices were obtained needs to be spelled out: the “product pricing” block on /marketing/intro renders server-side as all placeholder ¥0, and the real figures are served by the official website’s own public interface. That interface path was read out of the website’s own JS chunk; it requires no login and no authentication, and it is exactly the one the page calls when rendering prices. Values actually retrieved from the interface: Free ¥0 (freeDays=7, 1 account / 3 models / 20 monitoring prompts / 1 project); Advanced ¥259/month, ¥622/quarter, ¥2176/year, 1 account / 40 prompts; Professional ¥659/month, ¥1582/quarter, ¥5536/year, 5 accounts / 120 prompts; Enterprise is unpriced, and the page reads “contact sales · customized on demand.” Technical support is divided by tier into WeCom customer service / 5×8 dedicated support / 7×24 dedicated support.
SOUGEO 搜极星 (source: https://www.sougeo.com/ ; the prices and engine methodology come from the FAQ accordion on that page, whose source file is https://www.sougeo.com/assets/js/home-DGZA7ud8.js , and which is also emitted as FAQPage structured data): its meta description describes it as “a multi-platform-neutral GEO insight platform,” with two product lines—a brand AI-visibility dashboard and “星盾验真” content-risk identification; the official FAQ states plainly that it “neither provides nor encourages” black-hat practices such as AI poisoning and rank manipulation. Its billing form differs from all the others—one-time billing by report tier rather than a monthly subscription: 9.9 yuan Express (the site labels it a limited-time special, list price 99 yuan, 9 insight metrics), 199 yuan Professional (23 metrics, single-day deep data, including the Express entitlements), and 399 yuan Dynamic Analysis (29 metrics, 7-day continuous tracking and competitor movement alerts, including the Professional entitlements). Another FAQ entry mentions an enterprise custom edition delivered by a dedicated analyst team within 24–48 hours and including human interpretation, unpriced.
Dageno (source: https://dageno.ai/pricing plus the home page and footer): an English-language site whose meta description describes it as a platform that connects GEO audits with autonomous agent execution, aimed at marketers; the home page is labeled “Trusted by 3,700+.” The pricing page has four tiers: Starter $79/month, Growth $199/month (labeled Most Popular), Scale $499/month, and Enterprise as Custom / Contact Sales; the top of the page says “All plans include a 7-day free trial,” the annual option says “Save with 15% off annual billing,” and all three paid tiers use “Start Free Trial” as their CTA. Quotas: Starter 24,000 credits / 1 project / 50 prompts; Growth 60,000 credits / 2 projects / 150 prompts; Scale 150,000 credits / 5 projects / 500 prompts; all three tiers have competitor monitoring Up to 10, frequency Daily, and unlimited countries and languages.
Evidence scope and complete source list (the verification time throughout is 2026-09-04 PDT)
The purpose of the table below is to let you curl every claim above back for yourself, and at the same time to spell out how far down we actually dug—where we retrieved only HTML, we cannot conclude on that vendor’s behalf that “it does not exist anywhere on the site.”
| Product | URLs actually crawled this time | JS bundle / structured data / interface searched? |
|---|
| AIDSO 爱搜 | https://www.aidso.com/ ; https://www.aidso.com/pricing (response body the same 341552 B as the home page) | No—2 HTML pages only |
| 透镜 GEO | https://geo.timus.cn/pricing ; https://geo.timus.cn/ | Prices and FAQ are printed directly in the HTML (including FAQPage structured data) |
| Zhituishidai GenOptima | https://zhituishidai.com/ (/pricing returns 404) | No—home-page HTML only |
| Newrank Geowise | https://geo.newrank.cn/ ; https://newrank.cn/ | No—HTML only; the logo carousel has only img1–img6 alt attributes, and text inside the images was not retrieved |
| Profound | https://www.tryprofound.com/pricing | Static HTML only; the monthly price is rendered by script and was not retrieved |
| ImpetaAI 慧辰 | https://impetaai.hcr.com.cn/ ; https://impetaai.hcr.com.cn/marketing/intro ; https://impetaai.hcr.com.cn/api/plan/showList | Yes—the interface path was read out of the website’s own JS chunk, and the interface requires no login |
| SOUGEO 搜极星 | https://www.sougeo.com/ ; https://www.sougeo.com/assets/js/home-DGZA7ud8.js | Yes—including the JS bundle and FAQPage structured data |
| Dageno | https://dageno.ai/pricing ; https://dageno.ai/ | No—2 HTML pages only |
(One verifiable summary statement: within this evidence scope, 5 of these 8 vendors print their prices on public pages of their official websites or on their public interfaces—透镜 GEO, Profound, Dageno, ImpetaAI, and SOUGEO; the other 3 require an inquiry—AIDSO, Newrank Geowise, and Zhituishidai’s standard tier and above. This is only a description of the state of public information and carries no judgment of better or worse.)
4. Our Own Row: Applying the Same Yardstick to Ourselves
First, the identifying information in full, so that you can verify it: YinJen 引见 from Suzhou ZhiMaHang Technology Co., Ltd. (zhimahang.com/yinjen): a desktop application whose external methodology covers 12 AI engines, with a Visibility Score and paragraph-level optimization recommendations. Creator plan ¥29.9/month, Team plan ¥99.9/month, 14-day free trial.
Then the data methodology, stated in full in one place:
- Planned dispatch: 45 fixed questions × 12 AI engines = 540 runs per round.
- Actual methodology: 495 successful interface returns per round = 11 engines × 45 questions (495 = 11 × 45). The missing engine is claude—the login session on our account expired, and it has produced nothing for 5 consecutive rounds. This is a problem with our own account and is unrelated to that engine’s service.
- 495 does not equal “495 valid answers received.” 495 is the “successful interface return” count, and it includes a small number of empty answers; in R8, for example, there were approximately 472 non-empty answers. The denominator in the table below is uniformly 495 (the successful-interface-return count), and that layer of discount is worth keeping in mind as you read it.
- How rounds are separated: by our internal
cycle_id, not by natural-day windows (separating by date would contaminate one round with another). - Time range: the natural-day intervals corresponding to each round were not exported together with this data pull, so this article does not list them, and we do not reconstruct them after the fact; to check by date, you would have to go back to the data source and look up the start and end timestamps by
cycle_id—and that database stores timestamps in UTC, so reading them in PDT requires -7 hours. - Matching rule: our system tags each round using one and the same brand-mention determination. Its implementation details (whether domains, English names, and misspelled variants are all counted, and how multiple mentions within a single answer are counted) were not verified item by item in this article, so these figures may be compared only between rounds within this table and cannot be aligned with any figure produced under an external methodology.
- Data pull and verification time: 2026-09-04 (PDT).
Within this measurement, the number of times we ourselves were mentioned (the denominator is 495 throughout):
| Round | R4 | R5 | R6 | R7 | R8 |
|---|
| Mentions of YinJen 引见 | 24 | 18 | 4 | 0 | 3 |
The above is a snapshot as of 2026-09-04 PDT; rerunning the same questions overwrites in place, and historical values cannot be looked up again. In other words, these figures can only be reported as the state at the moment of the data pull; they cannot be treated as a permanently traceable historical ledger.
R7 is 0. We do not intend to hide that number—a set of figures published only when it looks good does not deserve to be treated as an example of methodology.
At the same time, the evidence boundary of this data set has to be stated: it can show only “who got written into the answers, for these 45 questions, these 11 valid engines, and these rounds.” It cannot show market share, it cannot show which product is better, and it cannot be used to derive what result you would get from your own campaign. Change the question set and the ordering could be completely different—which is exactly why all GEO data has to be read methodology-first.
5. How to Recalculate It Yourself: Pin Down Six Fields and Anyone’s Data Can Be Verified
You do not have to trust the conclusion any vendor gives you, ours included. The cost of GEO measurement is not high, and you can perfectly well run it yourself. The key is to pin down the following six fields before you start pulling data:
- Question set and denominator: fix a batch of questions first, 10–30 of them, write down the total count, and distinguish between “number planned for dispatch,” “number of successful interface returns,” and “number of non-empty answers”—these three numbers are often different, and mixing them will fool you.
- Engines and access endpoints: list item by item exactly which ones, and whether via the web or the APP. Do not record only a total—six months later, you yourself will not remember which 6 the “6” stood for.
- How rounds are separated: use a round identifier rather than a natural-day window. If a batch of tasks that spans natural days is separated by date, data from two rounds gets stirred together.
- Matching rule: what counts as “mentioned”? Do the full brand name, the short form, the domain, and the English name all count? Does a plain-text mention with no link count? Do misspelled variants count? If one answer mentions you three times, is that 1 or 3? Leave this unpinned, and two people counting the same batch of answers will arrive at two different numbers.
- Time zone: this is the easiest one to get wrong. Many systems store timestamps in UTC, and cutting windows by local time shifts everything by several hours.
- Verification time and snapshots: in some systems, rerunning the same question overwrites in place, and historical values cannot be looked up again. To keep evidence, save a separate snapshot of the results at the end of each round and record the verification time.
The recomputation procedure itself is simple: same questions, same engines, run within as short a time window as possible → count once using the matching rule above → record “mentions / successful interface returns” → repeat next round. Single-round figures fluctuate a great deal; wait until you have at least three rounds before looking at a trend.
When the sample is too small and the numbers jump around, there are only four things you can do: add questions, add engines, raise the dispatch frequency, and merge multiple rounds to lengthen the period—all four are doing the same thing, making the denominator bigger. Note that adding questions makes the rounds before and after incomparable (once you change the question set, you should start a new round and note “question set changed”), while raising the frequency changes what “a round” means; both have to be written into the methodology.
Finally, turn this checklist around and it becomes a template for questioning any GEO software (and any figure they publish):
- What is the denominator? How many runs were planned, how many interface returns succeeded, how many answers were non-empty?
- How are rounds separated? By natural day or by round?
- What time range does it cover? What time zone?
- How does the matching rule define “mentioned”?
- When was this data verified? Does a rerun overwrite it? Does it still hold now?
- Is the engine count on the official website a count of models, of engine endpoints, or of platforms? Which tier does it apply to? Is there a complete named list?
Ask those six questions and most incomparable numbers will reveal their methodology differences on their own.
6. Boundary Statement
This article is a snapshot copied down on 2026-09-04 (PDT) from the pages listed in the Section 3 list. Prices get adjusted, engines get added and dropped, and pages get redesigned—please take what you see when you verify it yourself as authoritative. Every “not seen” and “not retrieved” in this article is accountable only for the URLs in that list and does not constitute a judgment that any vendor “does not have it anywhere on their site.”
This article makes no promise about the results of any tool: no GEO software can guarantee that you will be indexed by AI engines, guarantee a ranking, or guarantee results within any given cycle—and our own product is no exception. YinJen does not promise “guaranteed indexing,” “guaranteed ranking,” or “guaranteed dominance.” This article does not score, does not rank, does not evaluate who is better or worse, and does not speculate about any vendor’s motives; if any of the information above does not match the current state of an official website, the official website prevails, and we welcome corrections so that we can update.
Further reading
