Snapshots

Overview

A snapshot is a recorded AI response to one of your prompts, captured at a specific point in time. Each snapshot is automatically analyzed for brand mentions, ranking position, sentiment, competitor references, citations, and content suggestions.

Snapshots are the raw data behind every metric in AEO Optima. Your visibility scores, trend charts, and competitor intelligence all derive from the snapshots you capture.

What Is a Snapshot?

When you capture a snapshot, AEO Optima sends your prompt to one or more AI answer engines and records the full response. The platform then runs automated analysis on each response to extract structured data.

Think of a snapshot as a timestamped photograph of how an AI model answers a specific question. By capturing snapshots regularly, you build a historical record that reveals how AI perception of your brand changes over time.

How to Capture a Snapshot

1. Navigate to Snapshots from the sidebar.
2. Click New Snapshot.
3. Select the prompt you want to send from the dropdown list.
4. Toggle the AI models you want to query. You can select one or multiple models per capture.
5. Click Capture.
6. Wait approximately 10 to 30 seconds for the AI models to respond and the analysis to complete.

Once the capture finishes, the snapshot appears in your list with analysis results immediately available.

Note: If no AI model options appear when creating a snapshot, you need to configure your LLM settings first. Go to Settings > LLM Configuration to enable the models you want to use.

What Gets Analyzed

Every snapshot is automatically analyzed for the following data points. No manual tagging or review is required.

Brand Mentioned

A simple yes/no indicator of whether the AI response contains a mention of your brand. This is the foundation of your Mention Rate metric on the Dashboard.

Rank Position

If the AI response contains a numbered or ordered list, AEO Optima identifies where your brand appears in that list. A rank of 1 means your brand was listed first. If your brand is not in the list, no rank is recorded for that snapshot.

Rank tracking is particularly valuable for prompts like "Best [product category]" or "Top [number] tools for [use case]", where position directly influences user perception.

Sentiment

Each response is scored for sentiment — positive, neutral, or negative — based on the language the AI model uses when discussing your brand or topic. This analysis considers the overall tone, specific word choices, and contextual framing.

For more detail on how sentiment is tracked and used, see Sentiment Analysis.

Competitors

The analysis identifies mentions of competitor brands within the AI response. This data feeds into the Competitor Intelligence features, including mention-share calculations and competitive benchmarking.

Citations

Many AI models include URLs or references to source material in their responses. AEO Optima extracts these citations and presents them as clickable links in the snapshot detail view. Citations help you understand which web pages AI models are drawing information from — and whether your own content is being cited.

Because captures are web-grounded (see below), citations on current snapshots are predominantly live sources the engine actually retrieved at capture time. On snapshots from before the grounding change — or the rare snapshot marked none — citations are URLs the model produced from its training memory, which is a different (and weaker) signal.

Web Grounding

Since August 6, 2026, every snapshot capture asks the AI engine to answer from a live web search, not from its training memory alone. Engines with built-in search (such as Perplexity Sonar) ground themselves; every other engine receives fresh search results attached to the request at capture time. Each snapshot records exactly how its answer was grounded in its capture provenance:

• native — the engine searched the web itself as part of answering.
• web_plugin — AEO Optima attached a live web search to the request.
• none — the answer came from the model's training memory alone. This is rare: bring-your-own-key direct connections, or when grounding is administratively disabled.

Grounded answers reflect what an engine can find about your brand today — the surface real users see in search-connected AI products. Ungrounded answers measure what the model remembers about your brand from its training data — a related but distinct signal.

Methodology note: snapshots captured before August 6, 2026 were grounded only on natively-searching engines. Trend comparisons that cross that date should expect a step change in citation counts and, on some engines, visibility metrics — the platform is measuring the live web from that date onward, not model memory.

Content Suggestions

Based on the analysis, AEO Optima may generate suggestions for improving your visibility in future responses. These suggestions are contextual and based on what was found (or not found) in the specific snapshot.

Reading Answers by Question

The Answers page opens on By question: the questions of one capture cycle down the left, and the answers to the question you pick on the right, one card per engine. Each question in the list shows its type with its code (P0 to P3), its topic, and "Named n/m": how many of the engines that showed an answer named you. An AI Overview or AI Mode answer that Google did not show, or one we could not read, is left out of both numbers, and the line's hover says how many were left out. On a phone the list becomes a menu above the cards.

Each card shows the engine and when the answer was captured, in your project's timezone, then its badges: Named #n when the answer named you at a position, Named without one, or Not named; the tone (Positive, Neutral or Negative) when a judge scored a verdict on an answer that named you; and Cited when one of its sources was a page on your own site. Hover over or focus the badges to read why they say what they say. Below them is the start of the answer, with your name and the competitors it named marked; Full answer shows the rest in place, and Open the answer page opens the answer's own page. The domains the answer cited follow, yours highlighted. When Google showed no AI Overview for the question, the card says so and nothing else. When an engine answered the same question more than once in the cycle (a re-run, another location or another language), its card shows the latest and says how many there were, with a link to all of them. The question you pick is kept in the address (?question=), so a link opens the same question.

All answers switches to the flat list of answers, newest first, with the period picker. Grid shows every question of the cycle at once: your questions down the side, the engines across the top. Each cell holds every answer that engine gave that question in the cycle, so a cell can hold several and says how many, and each answer reads as it does in the list: named or not named, its tone, whether your own site was cited. Every answer opens its own page, and every question links to all answers to that question.

Choose the cycle with the cycle picker (cycles are numbered from your project's first capture, C1 onwards), or step to the one before or after. A project with no capture rhythm yet shows its latest capture day instead of a cycle. By question and Grid show one cycle, not a period, so the period picker is hidden in them. The view is kept in the address (?view=detail, ?view=list or ?view=grid), so a link opens the same view. A link to all answers to one question, or a link that carries its own dates, opens the list.

How many questions a page holds

By question and Grid page by question. A page holds at most 50 questions, and fewer when the cycle is busy: 1,000 divided by the answers of the busiest question in the cycle, rounded down. That is because a page's answers are read in one read, and one read returns at most 1,000 answers, so the page is sized for its answers to fit in that one read. A cycle whose busiest question has 23 answers pages 43 questions at a time (43 × 23 = 989). The line under the page says how many questions a page holds and why. If the busiest question cannot be counted, the page holds 25 questions and says so.

When a page does not fit one read

If a page still holds more answers than one read returns (the busiest question could not be counted, one question alone has more than 1,000 answers in the cycle, or answers were captured while the page loaded), the page says so above the questions. The questions at the end of that page then show only some of their answers, or none. While a page is cut like this, By question says "Named at least n", because the count is a floor, and in the Grid an empty cell reads "Not read" instead of a dash: an answer that was not read is not an answer that was not captured. A dash always means the engine gave that question no answer in that cycle.

Viewing Snapshot Details

Click any snapshot in the list to open the full detail view. The detail view includes:

• Full AI Response — The complete text returned by the AI model, exactly as it was generated.
• Analysis Summary — A structured breakdown of brand mention, rank, sentiment, and competitors detected.
• Citations — All URLs referenced in the response, displayed as clickable links for easy verification.
• Content Suggestions — Actionable recommendations derived from the snapshot analysis.
• Capture Provenance — a disclosure line stating exactly how the answer was collected: captured natively by AEO Optima's own browser over a residential connection with the capture country named (Google AI Overview / AI Mode), via the provider's own API (chat LLMs), or via a SERP vendor's API (connector-based captures). An answer is labelled as routed through a residential connection only when that connection actually carried it. If the residential connection fails and the answer is collected directly instead, the line says the location was given as a hint in the prompt, with no proxy.

Understanding Snapshot History

Snapshots accumulate over time and form the basis of your trend data. A few things to keep in mind:

Frequency Matters

The more regularly you capture snapshots, the more granular your trend data becomes. Daily snapshots give you a clear picture of week-over-week changes. Weekly snapshots are sufficient for slower-moving categories.

AI Responses Vary

AI models are non-deterministic — the same prompt can produce different responses at different times. This is expected behavior, not a bug. It's one of the primary reasons regular snapshot capture is valuable. Patterns emerge over dozens of snapshots, not from any single response.

Multi-Model Comparison

When you capture snapshots across multiple AI models for the same prompt, you can compare how different engines represent your brand. One model may consistently mention your brand while another does not. This insight helps you understand where your visibility is strongest and where it needs improvement.

Tip: Set up automated snapshot schedules to ensure consistent data collection without manual effort. Regular captures are the key to reliable trend analysis.

Cost and Usage Tracking

Every snapshot generates a usage record that tracks:

• Prompt tokens — The size of the input query sent to the AI model
• Completion tokens — The size of the AI model's response
• Cost — Calculated automatically based on model-specific pricing from the model registry

These records are visible on the Usage page, where you can see breakdowns by model, provider, and time period. This provides full cost transparency for every capture.

Snapshots and Other Features

Snapshots feed data into several other areas of the platform: