The short answer
Tracking tools answer what happened. Analysis answers what to do about it, and it is where most AEO programmes quietly fail: a dashboard full of trend lines, a team with no idea which page to fix first.
The four analyses that convert data into decisions are citation-source mix, competitor gap, prompt-level loss diagnosis, and sentiment. The tools that support them best are the trackers with real citation-source reporting: Peec AI from $95/mo is the strongest per dollar and what we run client analysis on daily; Profound matches the depth at enterprise packaging; Otterly.ai covers the basics. The full market is in our best AEO tools guide.
But the honest headline is that the tool is the smaller half. The same dashboard produces a shrug or a strategy depending on the questions you put to it. This guide is the questions.
Analysis 1: citation-source mix
The first analysis to run on any AEO dataset: whose pages are the engines actually citing in your category? Group every cited URL from your tracked prompts by domain, then by type: your site, competitor sites, editorial and listicles, review platforms, UGC and forums, reference sites.
The result is usually the most clarifying chart in the programme, because the mix is not what teams expect:
The action output: rank the third-party domains by citation frequency, mark the ones where you are absent, and you have a prioritized outreach and placement list. This single analysis usually reorders an entire content roadmap.
Analysis 2: competitor gap
Run the same mention and citation math for every competitor on identical prompts, then look for the asymmetries rather than the totals:
- Prompts a single rival owns. Consistent first-position presence on a commercial prompt is rarely luck; open the citations and find which source is carrying them. That source is your target, either to appear on it or to displace it with something the engines like better.
- Prompts everyone loses. Nobody from your competitive set appears, and the answer cites generic editorial. These are open goals: often one strong, genuinely useful page can become the default citation for the whole prompt family. Our own experience ranking for tool-comparison prompts came from exactly this pattern.
- Engine splits. You win a prompt on Perplexity and lose the identical prompt on ChatGPT. Since the engines retrieve from different indexes, the diff between their citation lists is a precise map of which sources each engine trusts, and where your coverage has a hole.
Analysis 3: prompt-level loss diagnosis
Aggregates find the problems; individual answers explain them. For each commercially important prompt you lose, read the full answer and classify the failure:
| Loss type | What you see | The fix runs through |
|---|---|---|
| Source gap | Engines cite roundups and review pages that omit you | Getting onto those pages, or creating the better source |
| Content gap | Your relevant page exists but is never retrieved | Answer-first restructuring, direct answers, schema |
| Entity gap | Engines never associate you with the category at all | Consistent entity signals, definitional third-party coverage |
| Narrative gap | You are mentioned but framed poorly or outdated | Correcting the sources the engines quote for the claim |
The reason this table matters: the four failures have almost disjoint remedies, and an aggregate score cannot distinguish them. Ten lost prompts might be one source gap repeated ten times (one outreach campaign fixes it) or four different failures (four different workstreams). You cannot know without reading the answers, which is why every analysis workflow we run keeps the raw responses one click away, and why tools that discard full answer text in favor of scores make analysis strictly worse.
Analysis 4: sentiment and narrative
Presence is not endorsement. When you do appear, how are you described? The failure modes worth catching: outdated facts (old pricing, discontinued products, stale positioning), lukewarm framing ("also worth considering"), and miscategorization that puts you in the wrong comparison set.
The analysis is simple: collect every sentence the engines say about you and read them together. Patterns jump out in minutes that no score surfaces. The remedy usually lives off-site: engines repeat what their sources say, so a persistent wrong claim traces back to a specific stale page the engines keep citing. Find it in the citations, fix or displace it, and the narrative follows. We covered a version of this remediation pattern in our Scrunch AI review, whose product is built around exactly this brand-representation problem.
What the tools support
| Tool | Source-domain analysis | Competitor benchmarking | Per-prompt drill-down | Sentiment |
|---|---|---|---|---|
| Peec AI ($95 to $795/mo) | Ranked cited-domain reports, the report we use most | Same prompts, full set | Yes, with answer context | Yes |
| Profound ($99/$399/custom) | Citation-source analysis, deepest at Enterprise | Yes | Yes | Yes |
| Otterly.ai ($29 to $489/mo) | Link citation analysis, lighter | Basic | Yes | Limited |
| Spreadsheet + manual reads | Whatever you log | By discipline | Always, you have the answers | By reading |
Two notes. Profound's analysis depth is real, but remember its $99 Starter analyzes ChatGPT only; the cross-engine analyses above need at least the $399 Growth tier. And the manual row is not a joke: at small prompt counts, reading every answer yourself is analytically superior to any dashboard, just unscalable.
From analysis to action list
The output of a good analysis cycle is not a report; it is a ranked to-do list. The mapping we use in client work:
- Source gaps on money prompts rank first: pitch, place, or build the missing citation source. Highest impact per unit of effort we know of in AEO.
- Content gaps rank second: restructure the never-retrieved page around direct answers before writing anything new.
- Open-goal prompts third: build the page the engines are waiting to cite.
- Narrative fixes as they surface, since single stale sources are usually quick to correct.
- Re-run the analysis after 30 days and let the trend arbitrate whether it worked.
Key takeaways
- Analysis is the layer that turns tracking data into decisions. Four analyses matter: citation-source mix, competitor gap, prompt-level loss diagnosis, and sentiment.
- Start with the source mix. Your own site casts 2 to 6 percent of the votes in our tracking; the ranked list of third-party cited domains is your real roadmap.
- Classify every lost prompt as a source, content, entity, or narrative gap. The fixes are disjoint, so the classification is the decision.
- Peec AI has the strongest analysis layer per dollar; Profound matches it at enterprise depth (from the Growth tier up); Otterly covers the basics. No tool asks the questions for you.
- End every cycle with a ranked action list and a 30-day re-check, or the analysis was decoration.
If you want the analysis done for you
Running these four analyses across engines, prompts, and competitors is most of what our audit engagements are: your dashboard data plus our reading of it, ending in the ranked action list rather than a deck of charts. If you have tracking in place and no clear next move, that gap is exactly what we fix. Book a strategy call.