Understanding the engines

Geotally runs every prompt against 6 engines: ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, and Google AI Overviews. They answer the same question differently, and those differences matter when you're deciding what to optimize.

This guide covers how each engine answers, why they diverge, and what we ask them.

Why these 6 engines

These cover the bulk of consumer and prosumer AI search traffic. Other engines exist (DuckDuckGo's AI assistant, You.com, Brave's Leo, Kagi's Assistant) but their traffic share is small enough that tracking them adds cost without proportional signal.

If a new engine reaches meaningful share, we add it — included on every plan at no extra cost, same as the current six.

You don't have to track all six. Each brand has an engine picker — if Copilot traffic is irrelevant to a client, turn it off for that brand and the dashboards stay focused on the engines that matter.

How we query each engine

We send your prompt as a single-shot user message. No follow-ups, no chain-of-thought instructions, no leading system prompt — we want an answer shaped the way a cold query gets answered.

Every engine is queried through an API — no scraping, no browser automation:

ChatGPT — OpenAI API with the web-search tool enabled, so answers are grounded in live pages the way browsing-enabled ChatGPT answers are, and we capture the URLs it pulls.
Claude — Anthropic API with web search enabled; citations come back as structured source annotations.
Gemini — Google API.
Perplexity — Perplexity API; retrieval is the product, so citations are native.
Google AI Overviews — retrieved via SerpAPI from real result pages.
Microsoft Copilot — emulated: Microsoft publishes no Copilot API, so we retrieve Bing results for your prompt and synthesize a Copilot-style answer from them (details below).

On model choice: we pin cost-efficient models on each API (this is a big part of why a 6-engine tracker can start at $19/mo) and validate against the consumer engines with regular smoke tests. The trade-off is honest — a pinned API model isn't bit-for-bit the consumer default — but which brands get named and which sources get cited track closely, and tracking those consistently over time is the job.

ChatGPT

The most-used AI engine, and the one that produces the most consistent answers across runs. ChatGPT tends to:

Lead with the largest, most established brand in a category
Name 3-5 alternatives with brief descriptions
Cite few sources unless retrieval is triggered
Maintain a fairly stable list of named brands week-over-week

What this means for your data: ChatGPT is the slowest engine to move. If you're new in a category, expect to wait 8-12 weeks before you appear in ChatGPT for most prompts. Once you appear, you tend to stay.

How to improve ChatGPT presence: ChatGPT's training data leans heavily on well-indexed, widely-linked content. Earning mentions on major industry publications, getting included in roundup posts, and being part of "X vs Y" comparison content all help. Direct technical SEO is a weaker lever here.

Claude

Conservative in its answers. Claude tends to:

Hedge more than the other engines ("depending on your needs, options include…")
Name fewer brands per answer, often just 2-4
Default heavily to the market leader
Refuse more prompts on content-policy grounds
Cite sources rarely, even when grounded retrieval is available

What this means for your data: Claude is the hardest engine to move. The bias toward the market leader is strong, and the smaller named-brand surface means there's less room to land. If you're a challenger brand, expect lower share-of-voice on Claude than on the other five.

How to improve Claude presence: Claude has fewer obvious levers. The strategies that help most are the slowest: long-running PR, getting into the comparison-content ecosystem (Capterra, G2, industry "vs" pages), and time. Don't chase Claude share at the expense of the other five engines.

Gemini

Google's engine, and the one most likely to surface answers grounded in current web content. Gemini tends to:

Cite real URLs more often than ChatGPT or Claude
Favor brands with strong organic search rankings (unsurprising — same parent)
Mix in conversational hedging with concrete recommendations
Update its named-brand list faster than ChatGPT when a new player gains traction

What this means for your data: Gemini moves with your SEO. If your Google rankings for category-level queries improve, your Gemini share-of-voice tends to follow within a few weeks. The two metrics aren't identical — but they're correlated more tightly than any other engine pair.

How to improve Gemini presence: Traditional SEO is the lever. Schema markup, content depth on category-level pages, earned links from sites with high domain authority. If you have a strong SEO program already, you'll see Gemini results sooner than the other engines.

Perplexity

The outlier — Perplexity is a retrieval-first engine. It answers questions by searching the web in real time and citing every source. Perplexity tends to:

Cite 4-10 URLs per answer
Surface smaller, newer brands more often (because it's reading current pages, not training data)
Vary answers more between runs (because the underlying search results change)
Be the fastest engine to reflect new content

What this means for your data: Perplexity is the most volatile and the most actionable. If you publish a high-quality piece on your category today, you can see Perplexity mention rate move within a week.

How to improve Perplexity presence: Be cited. Perplexity's answer is built from the pages it retrieves, so the question becomes: which pages does Perplexity find when it searches your category? The cited URLs panel in your dashboard answers exactly that. Get mentioned on those pages — guest posts, sponsored mentions, earned roundups, partner content — and Perplexity will start citing you alongside them.

Why the engines disagree

Three reasons:

Different training data cutoffs. ChatGPT and Claude have static training data with quarterly-ish refreshes. Gemini and Perplexity supplement with live retrieval. A brand that launched last month doesn't exist in ChatGPT's worldview yet, but Perplexity sees it.
Different default behaviors on listing. Claude is comfortable naming two options and stopping. ChatGPT will name five. Perplexity will name everything it found in the top retrieval results. The surface area to land in is different per engine.
Different incentives. Each engine optimizes for different things — ChatGPT for plausibility, Claude for caution, Gemini for grounding in Google's index, Perplexity for citable retrieval. Those incentives produce different answers to the same question.

What "coverage" means

A brand has "coverage" on an engine when it shows up consistently in answers to its target prompts. Coverage isn't binary — it's a spectrum from "never mentioned" to "leading the answer most of the time."

Strong coverage on one engine doesn't guarantee coverage on another. We see customers with 60% mention rate on Perplexity and 5% on Claude all the time. The work to fix the Claude gap is different from the work that earned the Perplexity coverage.

Trying to be evenly covered across all 6 engines is usually the wrong goal. Some engines will always lag (Claude for challengers, ChatGPT in fast-moving categories). Focus on the engines where movement is possible and your customers actually use.

Google AI Overviews

Google AI Overviews are the AI-generated summary that appears above some Google search results. Google doesn't expose a direct API for them, so we retrieve them via SerpAPI. The overviews are inconsistent (they appear for some queries, not others; they vary by user, region, and signed-in state), and we record whether they appeared, whether your brand was named, and the cited URLs.

Microsoft Copilot

Microsoft does not publish a public API for Copilot. We emulate the Copilot engine by retrieving Bing organic search results for your prompt via SerpAPI and synthesizing a Copilot-style answer from those results using OpenAI GPT-4o. We do not transmit your prompt or any data to Microsoft. If you do not want your prompt processed through this pathway, exclude Copilot from your engine selection.

Engine failures and partial snapshots

Engines fail. Rate limits, timeouts, content-policy refusals, occasional outages on the provider side. When an engine fails:

The snapshot still finalizes with the engines that succeeded.
The failed engine is marked on the snapshot timeline.
Trend calculations exclude failed engines for that point in time so your rolling averages don't get distorted.

If one engine is consistently failing your prompts (more than 20% of runs), the dashboard surfaces a banner. Common cause: the prompt is triggering content-policy refusals on one engine but not others. Rephrase the prompt or drop it.

Choosing where to invest

If you have limited GEO budget, the rough order of leverage is:

Perplexity. Fastest to move, most actionable, citation-driven.
Gemini. Moves with your SEO, so any SEO investment compounds here.
Google AI Overviews + Copilot. Both leverage live web retrieval (Google's index for AIO, Bing's for Copilot), so they move with the same SEO and PR work that helps Gemini and Perplexity.
ChatGPT. Slower to move, but the largest audience by far, and stable once you're in.
Claude. Hardest to move directly; usually moves as a lagging indicator of broader category presence.

Most teams should focus on the retrieval-driven engines (Perplexity, Gemini, Google AI Overviews, Copilot) for the first six months. ChatGPT and Claude will follow as your category visibility grows.