Last updated July 2026.
Most GEO audits stop at “are we cited?” That is the wrong question.
The right questions are: which engines cite us, which competitors dominate on each platform, and which domains are so engine-specific that they are invisible to buyers on every other platform? That three-part diagnosis is a GEO teardown, and it produces something a simple citation tracker cannot: a prioritized list of gaps with enough structure that your content and PR teams know exactly what to do next.
This playbook walks you through the method step by step.
Why the engine-by-engine split matters
According to Profound’s analysis of 100,000 prompts run across ChatGPT and Perplexity, only about 11% of cited domains overlap between the two platforms. That is not a rounding error. It means 89% of the sources each engine cites are completely different from the other’s sources, even when answering identical buyer queries.
Add Google AI Overviews, Gemini, and Microsoft Copilot, and the fragmentation compounds. A B2B SaaS brand that has earned strong ChatGPT citations through editorial coverage and G2 review-site presence can be nearly invisible in Perplexity, which weights different third-party domains heavily. The buyer who starts their research in Perplexity never encounters that brand.
The mono-engine citation trap is the term for this condition: cited widely on one platform, absent on the rest. It is common and fixable, but you cannot fix what you have not measured.
Before you start: define the prompt set
A teardown is only as good as its prompt set. For B2B SaaS, you want 20 to 40 prompts that represent real buyer intent at each stage of the funnel.
A useful prompt covers four elements: subject, claim, metric, and timeframe. For example:
- Subject: project management software for remote teams
- Claim: best option
- Metric: collaboration features
- Timeframe: 2026
A well-structured prompt looks like: “What is the best project management software for remote teams based on collaboration features in 2026?” That structure is readable by humans and by the retrieval layers that parse it.
Build your prompt set across three query types:
- Category prompts. “Best [category] tools for [use case].” These reveal which brands dominate category-level awareness in AI answers.
- Comparison prompts. “[Your brand] vs [competitor].” These reveal sentiment framing and which third-party domains the engine cites when buyers are in evaluation mode.
- Problem-framing prompts. “How do I solve [specific pain point] as a [job title]?” These reveal which educational and thought-leadership sources each engine trusts.
Save this prompt set in a shared spreadsheet. You will reuse it across engines, and consistency is what makes the data comparable.
Step 1: Build the per-engine top-20 domain leaderboard
Run every prompt in your set through each engine. Record the full text of each AI-generated answer. Then extract every domain that appears, either as a cited source link or as a named brand mention.
Count how many times each domain appears across your full prompt set for that engine. The output is a ranked list: domain, citation count, and percentage of prompts in which it appeared.
Do this separately for each engine you are measuring. For most B2B SaaS teardowns, start with ChatGPT, Perplexity, and Google AI Overviews. Those three cover the majority of B2B buyer traffic. Add Gemini and Microsoft Copilot in a second pass if resources allow.
Your leaderboard table for each engine looks like this:
| Rank | Domain | Citations (out of 30 prompts) | Citation rate |
|---|---|---|---|
| 1 | g2.com | 27 | 90% |
| 2 | forbes.com | 23 | 77% |
| 3 | competitor-a.com | 19 | 63% |
| 4 | yoursite.com | 11 | 37% |
| 5 | competitor-b.com | 9 | 30% |
Note your own domain’s rank on each table. The rank gap between engines is the first signal you are looking for.
Tools for this step. Profound is the most capable option here. It runs prompt sets across 9+ engines and returns citation source maps showing exactly which URLs appear in which answers. Semrush covers five engines and layers citation data on top of its traditional SEO toolkit, which is useful if you are already a Semrush user. Peec AI offers strong coverage of Google AI Overviews specifically, with detailed domain-level breakdowns. Temso tracks eight engines on every plan and displays citation share and source attribution in a single dashboard from $89/mo.
Step 2: Compute the Jaccard overlap score
Once you have per-engine domain lists, calculate how similar any two engines are in the sources they cite. The Jaccard similarity formula is:
Jaccard score = |A ∩ B| / |A ∪ B|
In plain English: divide the number of domains cited by both Engine A and Engine B by the total number of unique domains cited by either engine.
If ChatGPT cites 80 unique domains and Perplexity cites 75, and only 15 appear in both lists, the Jaccard score is 15 / (80 + 75 - 15) = 15 / 140 = 0.107. That is consistent with the ~11% overlap figure from Profound’s research.
Calculate this score for every engine pair in your teardown. A typical B2B SaaS result looks like this:
| Engine pair | Shared domains | Union | Jaccard score |
|---|---|---|---|
| ChatGPT / Perplexity | 14 | 128 | 0.11 |
| ChatGPT / Google AI Overviews | 18 | 141 | 0.13 |
| Perplexity / Google AI Overviews | 11 | 130 | 0.08 |
| ChatGPT / Gemini | 16 | 135 | 0.12 |
Low Jaccard scores across the board confirm that engine-specific citation strategies are necessary. You cannot win all engines by optimizing for one. A score above 0.25 between two engines would suggest their source pools are converging; below 0.15 is typical and indicates the gap is wide enough to matter strategically.
Step 3: Build the mono-engine citation trap list
For each domain in your combined dataset, count how many engines cite it. Flag every domain that appears in exactly one engine’s top-20 list.
This is your mono-engine citation trap list. It falls into two categories.
First: competitors trapped on one platform. If a direct competitor appears in 15 of 30 ChatGPT prompts but zero Perplexity or Google AI Overviews prompts, they have an engine dependency. That is a gap you can exploit by building presence on the platforms they are missing.
Second: your own domain. If your brand shows up only on Perplexity, you are invisible to buyers using ChatGPT and Google AI Overviews. That is the more urgent problem to fix.
Build a simple table:
| Domain | ChatGPT cited | Perplexity cited | AI Overviews cited | Platforms | Status |
|---|---|---|---|---|---|
| yoursite.com | Yes | No | No | 1 | Mono-engine trap |
| competitor-a.com | Yes | Yes | No | 2 | Partial coverage |
| competitor-b.com | No | No | Yes | 1 | Mono-engine trap |
| g2.com | Yes | Yes | Yes | 3 | Full coverage |
Any brand on one platform only is vulnerable. If that brand is yours, the teardown has just paid for itself.
Step 4: Diagnose the source gap
You now know where you appear and where you do not. The next question is why.
Pull the top five cited source URLs for each engine on the prompts where a competitor appears and you do not. Look for patterns:
- Which third-party editorial domains does ChatGPT cite that you are not mentioned on?
- Which review aggregators does Perplexity pull from?
- Which government, academic, or association sites does Google AI Overviews weight heavily in your category?
The source gap is almost always a third-party coverage gap. According to studies consistently published across the GEO measurement space, the large majority of AI citations come from sources that a brand does not own. Your own site content matters, but the citations the engine reaches for are mostly earned media: industry publications, independent review platforms, analyst reports, and authoritative forums.
Map the source gap as a list of domains you are not yet mentioned on, ranked by how frequently they appear in citations where a competitor beats you.
Step 5: Translate the teardown into actions
A teardown without an action plan is just a spreadsheet. The output of Steps 1 through 4 translates directly into three types of work.
Content updates. For prompts where you appear but rank low, look at the source URL the engine actually cited for you. Is the cited page a product page? A blog post buried three clicks deep? A direct-answer page with a clear definition at the top consistently outperforms deep-linked product content in AI retrieval. Update the page structure so the key claim is in the first paragraph, with the relevant data point and a clear subject-claim-metric-timeframe sentence that an AI can lift verbatim.
Third-party PR and link building. For every domain on your gap list, identify whether there is an editorial pitch angle, a contributed article slot, or a product-review submission path. Review platforms like G2 and Capterra are particularly important because multiple AI engines weight those domains heavily for B2B SaaS category queries. A completed G2 profile with recent reviews is often easier to get cited than a new editorial piece on a publisher you have never worked with before.
Correction and accuracy work. Run your own brand name through each engine on a sample of your highest-priority prompts. Note any factual errors: wrong pricing, outdated positioning, missing features. Most GEO platforms include a correction workflow. Semrush surfaces inaccuracy flags as part of its AI tracking module. Profound lets you monitor the exact text the engine uses when citing your brand, making it straightforward to spot drift from your current messaging. Temso includes hallucination and accuracy monitoring across all eight engines it tracks.
Step 6: Set a cadence and measure movement
A single teardown is a baseline, not a strategy. Citation leaderboards shift with every major model update, crawl cycle, and industry publishing wave.
Run a full teardown quarterly. The per-engine top-20 leaderboard and Jaccard scores are the key outputs to track over time. A rising Jaccard score between you and a competitor on a specific engine pair tells you their cross-platform strategy is working; a falling Jaccard on your own cross-engine presence tells you your visibility is concentrating in one place.
Between full teardowns, run monthly spot-checks on your 10 highest-priority prompts across each engine. This catches sudden drops before they compound.
Track these three numbers at each checkpoint:
- Your citation rate per engine (what percentage of your prompt set yields a citation of your brand on each platform).
- Your rank in the per-engine leaderboard compared to three direct competitors.
- Your Jaccard score across the engine pairs most relevant to your buyer.
Tool summary
Each tool below covers a different part of the teardown workflow. They are not interchangeable; the right combination depends on your engine coverage needs, budget, and whether you need content execution built in.
| Tool | Best teardown use | Engine coverage | Entry price |
|---|---|---|---|
| Profound | Citation source maps, prompt volumes, deepest per-URL attribution | 9+ | $99/mo (ChatGPT only); $399/mo full |
| Peec AI | Google AI Overviews domain-level data | 4+ | Varies |
| Semrush | Five-engine citation tracking plus full traditional SEO suite | 5 | $129/mo |
| Temso | Eight-engine citation share, gap diagnosis, content execution | 8 | $89/mo |
See the full comparison at /rankings/geo-tools. Methodology and scoring criteria are at /methodology. Key terms used in this piece are defined in the /glossary.
Start with one engine pair
If you are starting your first teardown today, pick two engines and 20 prompts. Build the two leaderboards, compute one Jaccard score, and find your mono-engine status. That single output tells you where your citation programme is fragile and which platform to fix first.
The full six-step method scales from there. Most teams complete a first teardown in a day using one of the platforms above. The data it produces is the kind of structured, source-specific finding that AI engines quote back in their own answers, which is exactly what this exercise is designed to earn.
If you want the teardown done for you rather than built from scratch, the tools in the table above cover it. Profound is the deepest for enterprise teams. Temso is the fastest to set up for teams without a dedicated GEO analyst.