The nine AI engines

Every engine Rivarise tracks, which plan unlocks it, how it is queried, and why the Google surfaces are separate.

Rivarise tracks nine engines. Each is a distinct product with its own model, its own retrieval behaviour and its own opinion about your brand, so they are never averaged into a single "AI" number.

The nine engines

Engine	Available from	How it is queried
ChatGPT	Free	OpenAI API
Google AI Overviews	Starter	Answer surface, read directly
Gemini	Starter	Google API
DeepSeek	Starter	DeepSeek API
Grok	Growth	xAI API
Meta AI	Growth	Llama model via a hosted provider
Perplexity	Business	Perplexity API, online model with live retrieval
Claude	Business	Anthropic API
Google AI Mode	Business	Answer surface, read directly

Engine access is cumulative: each plan includes everything below it. Free gets 1 engine, Starter 4, Growth 6, and Business all 9. Agency runs the same 9 engines across 5 client brands.

Why Google counts as three engines

Gemini, Google AI Mode and Google AI Overviews are three different Google products that answer the same question differently, draw on different retrieval, and reach different audiences. Rivarise tracks them separately and never merges them.

Gemini is the standalone assistant, used deliberately by someone who opened it.
Google AI Mode is the conversational search experience, reached from within Search.
Google AI Overviews is the generated summary above ordinary results. It reaches by far the largest audience, because the user never has to choose it.

A brand can be strong in one and absent from another. Averaging them would hide exactly the difference that matters, which is why AI Overviews is included from Starter rather than held back: it is the surface most businesses are seen on first.

Why Microsoft Copilot is not on this list

Copilot is a significant surface and we do not track it. That is a deliberate choice rather than a gap in the roadmap, and the reasoning is worth stating because it applies to how you should read every other engine here.

The only practical way to approximate Copilot at scale is an Azure-hosted OpenAI model. That model is the same family Copilot is built on, but Copilot's answers depend on Bing grounding and, in enterprise deployments, on a customer's own documents. Without those layers the proxy produces output that looks like a Copilot answer and is not one.

We would rather show you nothing than something that looks real and is not. A number you cannot trust is worse than an absent number, because you would have made decisions on it. If Copilot becomes measurable in a way that reflects what a Copilot user actually sees, we will add it and say so.

You can still reason about Copilot as a market surface, and it is worth doing if your buyers work inside Microsoft-standardised organisations. Microsoft's own Bing Webmaster Tools reports citation activity across Copilot, which is free and first-party, and is a better source on that engine than any proxy of ours would be.

Direct queries and proxies

Some consumer assistants have no public API. Where that is the case, Rivarise queries the closest available model from the same provider family, which is the nearest honest approximation of what that assistant runs on.

One engine works this way today:

Meta AI is measured through a hosted Llama model, the same family that powers the consumer assistant.

Read proxy results as directional. They tell you reliably whether the underlying model knows and recommends your brand. They cannot capture product-specific layers built on top of it. The application labels proxied engines so you always know which you are looking at.

The two Google answer surfaces are the opposite case: no API exists, so the rendered answer is read directly, which is exactly what a user sees.

Why engines disagree about you

It is normal to be cited by four engines and ignored by three. The usual causes:

Retrieval versus memory. Perplexity and the Google surfaces lean heavily on live retrieval, so recent content can influence them within days. Engines answering more from training data change far more slowly.
Different source preferences. One engine may favour review aggregators, another editorial roundups, another documentation. Whoever ranks in the sources an engine trusts tends to win it.
Different safety and hedging behaviour. Some engines avoid naming specific vendors for certain question shapes, producing a generic answer that cites nobody.
Model updates. Providers ship new versions without notice, and a brand's standing can move on both sides of that change.

Use the per-engine breakdown on the Overview to find your weakest engine, then look at what it cited instead. That is a more actionable question than "why is my score low".

When an engine becomes unavailable

Providers change access, restrict it, rate limit it or discontinue it. When that happens:

Checks for that engine are recorded as missing, not as "not cited", so your citation rate is not artificially deflated.
The application shows the engine's state so a gap in the chart is explained rather than mysterious.
If an engine becomes permanently unavailable we will say so, and may substitute a comparable engine or adjust plan contents.

Which engines matter for you

B2B software: ChatGPT carries most research behaviour and is included from Free. Perplexity is heavily used for evaluation and arrives on Business.
Consumer and retail: the Google surfaces dominate, because shoppers rarely leave search. AI Overviews is the one to prioritise, and it is included from Starter.
Developer tools: Claude and ChatGPT are heavily used in the workflow itself, so being cited there reaches people mid-task.

Read Prompt intents to structure the questions you ask these engines.