Generative engine optimization (GEO) is the practice of measuring and improving how a brand appears in answers written by AI systems. On 10 October 2026 we checked which AI engines nine GEO vendors say they monitor. Eight list ChatGPT, Claude, Perplexity and Gemini, though some engines sit on higher plans or cost extra. Semrush does not mention Claude.
So most tools "track" all four. But each engine name hides choices about how the tool reaches the engine, and vendors document them unevenly. Before you buy, get each engine's configuration in writing. For pricing and metric definitions, see our comparison of nine GEO platforms.
What does a tick in the engine list leave out?
At least six choices sit behind each engine name:
- Surface. The consumer interface is the app or website a person uses, such as chatgpt.com. The developer API (application programming interface) is the paid access that software uses.
- Model or mode. A named model version, or whatever the app picks by default.
- Web search. Whether the engine looks things up online before answering, which is what produces linked sources.
- Country. Where the prompt appears to come from, and whether the engine is available there.
- Refresh cadence. How often the tool reruns the same prompt.
- Source capture. Whether the tool stores the web pages behind an answer. Some tools separate sources, every page the engine consulted, from citations, the pages linked in the answer.
"Gemini" in these lists means Google's chatbot. Google's AI Overviews and AI Mode usually appear as separate engines.
Facts below come from each vendor's public pages; we mark vendor claims that cannot be checked from outside. Not documented (ND) means we found nothing public, not that a capability is missing. We found no independent per-engine audit of any of these tools.
In the documentation we checked, vendors differ most on surface, and Claude shows it most clearly.
Why does Claude differ most between tools?
Three vendors describe or label their Claude feed as the developer API, even where they use the consumer interface for other engines.
Otterly.AI collects answers through public websites, but says "we only use API for Claude tracking" (data collection). That feed uses an unnamed Sonnet model "with web search capabilities" (supported engines).
Ahrefs Brand Radar tracks Claude "via API (Claude Sonnet 4.6 with web search)", for custom prompts only (help article, 3 September 2026).
Peec AI lists "Claude Sonnet 4 – API" on its Enterprise plan, without saying whether web search is on (pricing).
Profound says the opposite: it "captures responses directly from the consumer browsing experience rather than from model APIs" (method statement). This covers every engine, Claude included, and Profound says it "reflects what real users actually see" (vendor claim). It names no Claude model or search setting.
Scrunch, Genezio, Rankscale and AthenaHQ do not document a Claude surface, as the table below shows.
The surface matters because the API and the app control search differently. In the API, web search is available only if the request includes the web search tool, and Claude then "determines when to search based on the prompt" (Anthropic docs). In the newer Claude app, there is no toggle: "Claude searches the web when it helps" (Claude help center). Either way Claude decides when to search; what differs is who controls whether search is available.
Location differs too. The API takes an approximate location as a request parameter (Anthropic docs), while the app may infer it from the user's IP address (Claude help center).
What are you getting for ChatGPT?
ChatGPT is described more consistently. Profound, Peec AI, Otterly.AI, Semrush and Ahrefs all collect it from the consumer interface, but the mode differs:
- Semrush analyzes "the latest model of ChatGPT in search mode" (knowledge base).
- Ahrefs uses "the default model", "without any stored user data or prior context" (help article).
- Peec AI runs logged out, so "the platform decides which model to use and whether it runs a web search" (docs).
- Rankscale lets users pick named GPT versions, with web search costing extra credits per run (changelog, 2 October 2026).
These results are not interchangeable. Leaving search to the app shows: Peec AI notes that ChatGPT "sometimes performs web searches and sometimes doesn't", and answers without search have no sources (chat docs).
Two tools also list a separate OpenAI engine: Peec AI's Enterprise plan has "GPT 5 Search – API" (pricing), and Genezio lists "GPT-5.4" next to ChatGPT (engine docs). Check which one a report draws on.
Countries differ as well. Otterly.AI marks ChatGPT as unavailable in its Czech Republic and Hong Kong markets (countries, 24 September 2026). Rankscale says ChatGPT prompts "can run in Czechia again" from 9 October 2026. Neither explains the cause.
If the best-described engine varies this much, the other two need closer questions.
What are you getting for Gemini?
For Gemini, the surface is mostly undocumented. Otterly.AI and Ahrefs use the public web interface. Semrush says the ChatGPT and Gemini data in its prompt database is "captured from real requests and not via any APIs of LLMs" (knowledge base). Genezio mentions "a newer Flash Lite generation" of Gemini models but not whether it uses the app or the API (engine docs).
Gemini is also often outside the base plan, for example as an add-on at Otterly.AI or Enterprise-only at Scrunch.
What are you getting for Perplexity?
For Perplexity, no vendor we checked documents a mode or model. Ahrefs and Otterly.AI use the public web interface.
Peec AI's surface is unclear. Its docs give only examples: one page names "Perplexity API" as a sample channel (model channels), while the API reference uses "perplexity-scraper" as a sample model ID (list model channels). Neither says which one Peec uses.
Scrunch does not say which platforms it reaches by browser and which by API, though its FAQ uses Perplexity as its example of simulated consumer interaction (FAQ). Semrush covers Perplexity only in a weekly report (knowledge base). The table brings the four engines together.
| Vendor | ChatGPT | Claude | Perplexity | Gemini |
|---|---|---|---|---|
| Profound | Consumer browser; mode ND | Consumer browser; mode ND | Consumer browser; mode ND | Consumer browser; mode ND |
| Peec AI | Interface, logged out; app picks model and whether to search. Enterprise also lists "GPT 5 Search – API" | Enterprise; labelled API, Sonnet 4; search ND | Add-on; surface unclear | Interface for most platforms (per docs); Gemini not specified |
| Otterly.AI | Web interface; mode ND | Add-on; API, Sonnet with search | Web interface; mode ND | Add-on; web interface |
| Scrunch | Method ND | Enterprise; method ND | FAQ example of simulated consumer use; mode ND | Enterprise; method ND |
| AthenaHQ | Listed; method ND | Listed; method ND | Listed; method ND | Listed; method ND |
| Rankscale | Chosen GPT version; search optional | Haiku 4.5 listed; surface ND | Listed; surface and mode ND | 3 Flash, 3.1 Pro listed; surface ND |
| Semrush | Latest model, search mode; prompt data from real requests, not LLM APIs | Not mentioned | Weekly report only; surface ND | Prompt data from real requests, not LLM APIs; mode ND |
| Ahrefs | Web interface, default model | Custom prompts only; API, Sonnet 4.6 with search | Web interface; mode ND | Web interface; mode ND |
| Genezio | Interface for most engines; this one ND. "GPT-5.4" listed separately | Interface for most engines; this one ND | Interface for most engines; this one ND | Interface for most engines; this one ND. "Flash Lite" models mentioned |
Checked 10 October 2026 against each vendor's public docs, help center, pricing page or changelog. Profound's and Genezio's method statements cover all engines at once. Plan limits appear where documented.
Surface and model are only part of coverage; the rest is when, where and what the tool collects.
How do refresh cadence, countries and source capture differ?
| Vendor | Refresh cadence | Per-engine country gaps | Sources and citations |
|---|---|---|---|
| Profound | Daily | ND | Citations tracked; per engine ND |
| Peec AI | Daily; Enterprise daily or weekly | Listed per channel, via API | Stored separately |
| Otterly.AI | Daily | ChatGPT: Czech Republic, Hong Kong | Citation links |
| Scrunch | Daily for 14 days, then every 72 hours | ND | Stored; some fields platform-dependent |
| AthenaHQ | ND | ND | Citation intelligence listed; per engine ND |
| Rankscale | User-set, hourly to monthly | ChatGPT blocked in Czechia until 9 Oct 2026 | Citations analysis listed; per engine ND |
| Semrush | ChatGPT, Gemini daily; Perplexity weekly | ND | ChatGPT citations mentioned |
| Ahrefs | Index monthly; custom prompts daily to monthly | Runs in the keyword's location | Linked URLs |
| Genezio | Daily | ND | Cited sources and fan-out queries; per engine ND |
Checked 10 October 2026.
Cadence can differ inside one product. Ahrefs refreshes its own prompt index monthly (methodology, updated 26 February 2026), while custom prompts can run daily (help article). Because Ahrefs tracks Claude only through custom prompts, a Claude trend can rest on daily runs while ChatGPT data in the index refreshes monthly.
Few vendors publish per-engine country gaps; Otterly.AI's table is the clearest. Peec AI lists "country codes that cannot be used with this channel", but only through its API (API reference).
Source capture depends first on the engine. In the Claude, OpenAI and Gemini developer APIs, the request must include a search tool, and the model then decides whether to use it. Only answers that searched return citations (Anthropic, OpenAI, Google). No search, no sources.
What vendors keep then varies. Peec AI stores sources and citations separately (chat docs). Otterly.AI captures "citation links" (data collection). Genezio stores the cited sources and the follow-up searches an engine runs, or "query fan-outs", for each conversation (conversation docs).
What this means: engines inside one tool are not always measured under the same conditions. At Otterly.AI, Claude answers come from the API with web search enabled, while ChatGPT answers come from the public website with search handling undocumented. Their cited-page counts are not like for like.
What should you ask a vendor about each engine?
For each engine, ask the vendor to state in writing:
- Surface: consumer interface or developer API, logged in or out.
- Model or mode: a named version, or the app's default.
- Web search: whether it is available, and whether the engine decides when to use it.
- Countries: where this engine cannot run.
- Cadence: how often its prompts refresh, and whether that depends on plan.
- Sources: whether sources and citations are stored and exportable.
- Plan: included, add-on or Enterprise-only.
Then spot-check a few prompts by hand in the same configuration.
Neither surface is simply the right choice. An API request fixes the model, whether search is available, and the location; a consumer feed shows the product people use, with that day's defaults. What matters is knowing which one you pay for, engine by engine.