Enterprise Competitive Analysis, in the Age of AI

How we measured. AI-citation figures come from a saved Ahrefs Brand Radar report run against a panel of 106 custom prompts, US, five engines, pulled 7 September 2026. Claude is excluded because that data source returned exactly 1.0 and 0.0 across every brand, which indicates a very small response count rather than a distribution. SERP and demand figures come from a live Ahrefs fetch for “seo competitive analysis”, US, top 10 organic positions, 9 September 2026. Ranking figures come from a single property over 90 days, 11 June to 8 September 2026, covering 1,063 matching query rows (Google Search Console, 2026). Citation counts are appearance counts on one panel on one day; they are not market share and they do not sum to anything. One domain, one window, one panel — a case study, not a law. Figures verified as of 9 September 2026.

What is enterprise competitive analysis in the age of AI?

---

Enterprise competitive analysis is the work of identifying which organisations capture the demand you are trying to capture. The classical version asks one question: who ranks above us. That question still matters, and on its own it now describes only part of the market.

A second question sits beside it: who gets cited when we do not. The two questions produce different lists, because ranking and retrieval are decided by different systems reading different signals. Ranking is settled once per query against a whole page. Retrieval is settled per answer, against whichever passage the engine can lift and quote.

Share of voice — how often your brand is mentioned in AI answers — is not the same measurement as share of model, the proportion of answers in your category that a model actually sources from your pages. A brand can be mentioned frequently and sourced rarely. You should measure both, and you should never report one as though it were the other.

Why does the ranked-above-you list disagree with the cited-instead-of-you list?

---

Because the two lists are built from different evidence. Ranking reflects link authority, relevance and query matching against a whole page. Citation reflects whether a specific passage was retrievable and quotable at the moment an answer was assembled.

On our own panel, run against 106 custom prompts across five engines on 7 September 2026, the most-cited domains were semrush.com at 85 citing responses, youtube.com at 83, ahrefs.com at 62, indexableai.com at 52 and reddit.com at 52 (Indexable, 2026). Two of the top five are user-generated platforms that no vendor in the category controls.

The single most-cited page in our category on that panel is a competitor's blog post at 20 citing responses; our own best-performing page recorded 9 (Indexable, 2026). A classical competitive report built on domain authority would not surface either fact, because neither is a ranking.

Start by accepting that the second list exists. Most enterprise competitive reviews still produce only the first one, then present it as a complete picture of the market.

Who captures the click that an AI answer creates?

---

Somebody else, in our case. We ran a traffic-ownership audit over the fifteen most-cited domains, classifying every citation by who owns the destination: our pages, a competitor's pages, a review platform, a marketplace, or an unaffiliated third party.

The result was blunt. Of 681 citing responses, our own pages held 52 — a 7.6% share (Indexable, 2026). Competitor-owned surfaces held 43.5% of the total (Indexable, 2026). Unaffiliated third-party surfaces held 48.9% (Indexable, 2026). The audit's severity rule fired at its highest level, because competitor surfaces captured more citations than owned pages did (Indexable, 2026).

The zero is the line to read twice. Review platforms captured 0.0% of citations on our panel, which means we are absent from the surface engines reach for on shortlist and comparison questions (Indexable, 2026).

A second check on the same fifteen domains disagrees, and the disagreement is the useful part. A citation-concentration audit returned 15 unique citing domains, a top-domain share of 12.5% and a diversity score of 92 out of 100, with concentration risk graded low (Indexable, 2026). Diversity measures fragility — how exposed you are if one citing source disappears. Ownership measures economics — who receives the visit the answer generates. A category can score 92 on diversity and 7.6% on ownership simultaneously, and ours does (Indexable, 2026). Apply both. Two limits apply to all of it: the pull returns the fifteen most-cited domains rather than a full census, and one day is one day.

Is “who is winning” the same answer on every engine?

---

No. On our panel and on one day, the answer inverts depending on which engine you ask. Our appearance rate ran 58.57% on Copilot, 28.09% on ChatGPT, 19.35% on Perplexity, 10.50% on Google AI Overviews and 1.17% on Gemini — a 57.4-point spread on one brand, one panel, one day (Indexable, 2026).

Running a competitor-gap comparison per engine sharpens it further. On Copilot, no tracked competitor sat above us. On Gemini, all of them did, with Semrush ahead by 89.47 points and Ahrefs by 22.22 points (Indexable, 2026). Same brands, same prompts, same date.

Claude is excluded from every figure above. That data source returned exactly 1.0 and 0.0 values across all brands, which indicates a very small response count rather than a real distribution, and publishing it would have manufactured a zero (Indexable, 2026).

Next, decide which engines your buyers actually use before you decide whether you are winning. A single blended number hides an inversion this large.

What does the SERP for this discipline reveal?

---

That the category is documented by its vendors and by almost nobody else. We pulled the live top ten for “seo competitive analysis” on 9 September 2026. Positions 2 and 5 are vendor product pages — seranking.com/competitor-traffic-research and moz.com/competitive-research. Positions 6, 7, 9 and 10 are vendor blog posts from Semrush, Siteimprove, the Digital Marketing Institute and Ahrefs (Ahrefs, 2026).

No independent framework page appears in the top ten. Every result either sells a tool or is published by a company that sells one. The AI Overview sits at position 1 and links out to SE Ranking and to YouTube (Ahrefs, 2026), which matches the citation pattern above: youtube.com is the second most-cited domain on our own panel at 83 citing responses (Indexable, 2026).

The demand shape says the same thing. The head term carries 1,700 monthly US searches at keyword difficulty 22, against a traffic potential of 28,000 (Ahrefs, 2026). Almost all of the available traffic sits in the long tail of method questions, not in the head term the vendors are fighting over.

Deconstructing the ranked set is sobering in one direction and encouraging in another. Among the four ranked pages for which Ahrefs reported a referring-domain count, the average was 147 (Ahrefs, 2026). Yet the Ahrefs blog post ranks tenth on 5 referring domains. Authority is doing the work, not the page.

Why is head-term position the wrong scoreboard?

---

Because on our own estate head-term position and actual performance have come apart completely. Over 90 days our head terms sat at average positions of 76.8, 88.7 and 89.0, while conversational question queries on the same property sat between 1.9 and 5.3 (Indexable, 2026).

One page makes the point on its own. Our automation page recorded 1,618 impressions at an average position of 3.8 over that window and zero clicks (Indexable, 2026). Position 3.8 is a competitive win by any classical scoreboard. It produced nothing.

The citation side diverges as well. In a separate scan of 223 pages, 18 ranked in Google's top ten and 3 of those were cited by an AI engine (Indexable, 2026). Our two most-cited pages ranked 30.5 and 28.1 (Indexable, 2026). Ranking did not predict retrieval in either direction.

Zero-click behaviour is the backdrop: 68.01% of US searches ended without a click between January and April 2026 (SparkToro, 2026). A scoreboard built on head-term position measures a surface where most of the outcome no longer happens.

What are the anti-patterns?

---

Reporting share of voice as though it were share of model. Mentions and citations are different measurements on different bases. Our appearance rate reached 58.57% on one engine while our pages held 7.6% of citations across the fifteen most-cited domains (Indexable, 2026).

Blending engines into one number. A 57.4-point same-day spread averages into a figure that describes no engine anyone actually uses (Indexable, 2026).

Publishing a zero you have not verified. The Claude data source returned 1.0 and 0.0 across every brand on our panel — a small-sample artefact, not a distribution (Indexable, 2026).

Tracking only the vendors you consider competitors. Two of our five most-cited domains are YouTube and Reddit (Indexable, 2026). Neither is a competitor and both take citations a competitive report would never look for.

Treating an average position as a result. Position 3.8 with 1,618 impressions produced zero clicks on our own estate over 90 days (Indexable, 2026).

How do you run this analysis yourself?

---

Seven steps, and the dataset behind them is a single citation export plus one live SERP fetch.

1. Step 1 — pull the domains cited on your category prompts. Use a fixed prompt panel, state the n and the date, and record which engines were included and which were excluded.
2. Step 2 — classify every cited domain by owner. Owned, competitor, review platform, marketplace, unaffiliated third party. The share held by your own pages is the number to report first.
3. Step 3 — compute concentration separately from ownership. A high diversity score and a low ownership share can coexist; they answer different questions.
4. Step 4 — split every figure by engine before you average anything. Then, and only then, decide whether a blended number is worth publishing.
5. Step 5 — audit any engine returning suspiciously round values. Exactly 1.0 and 0.0 across all brands means too few responses. Exclude it and say so.
6. Step 6 — pull the live top ten for your head term and read the publishers. Count how many results are published by companies selling a tool in your category, and schedule an independent-method page if the answer is most of them.
7. Step 7 — join your ranking data to your citation data at page level. Implement it as a standing report. The pages that rank and are not cited are your working list.

If step 2 returns an owned share under half, your competitive problem is a distribution problem before it is a content problem. You can fix distribution faster than you can fix authority.

In summary

---

Enterprise competitive analysis now has to answer two questions instead of one, and the second one is harder to hear. Who ranks above us is a question about pages. Who is cited when we are not is a question about surfaces, and the surfaces include platforms nobody in the category owns.

Three numbers frame the work. Our own pages held 7.6% of the citations across the fifteen most-cited domains on our panel (Indexable, 2026). Review platforms held 0.0% (Indexable, 2026). And the same brand on the same day scored 58.57% on one engine and 1.17% on another (Indexable, 2026).

None of those is visible on a rank tracker. All three change what you would fund next quarter: a review-platform presence rather than another head-term push, an engine-by-engine scoreboard rather than a blended one, and a distribution plan that includes surfaces you do not own.

The cheapest next step is step 2 above. Export the domains cited on your category prompts, classify each one by who owns it, and compute your own share. If it comes back under half, you have found the gap your ranking report has been hiding.