Enterprise SEO, in the Age of AI

How we measured. Google Search Console, property indexableai.com, 90 days from 11 June to 8 September 2026, 1,063 matching query rows. AI citation counts from Ahrefs Brand Radar across seven engines, pulled 9 September 2026. Search Console positions are impression-weighted averages blended across device and country; treat positions beyond page one as directional. This is one domain over one window — a case study, not a law.

What is enterprise SEO in the age of AI?

---

Enterprise SEO is the practice of making a large organisation's information findable, retrievable and quotable at scale. The scale part is unchanged: thousands of pages, competing internal owners, a release train you do not control. What changed is the destination. Content is no longer only competing for a click on a results page. It is competing to be retrieved as source material by a system that answers the question itself.

That gives the discipline two success conditions. The first is familiar — rank well enough to be seen. The second is newer — be structured, current and specific enough that a retrieval system can lift a passage out of the page and attribute it. These are correlated but not identical, and at enterprise scale the gap between them becomes large enough to count.

You can test this on your own estate in an afternoon, and the rest of this piece shows you how. Start by accepting that the two outcomes need two measurements. Everything else follows from that.

What stayed the same, and what AI changed

---

The honest answer is that the fundamentals held and the scoreboard broke. Crawlability, architecture and authority still decide whether you compete. What changed is that being indexed stopped implying being retrievable, and clicks stopped being a complete measure of reach.

The unchanged column is why enterprise SEO is not being replaced. The changed column is why the reporting most teams present to their executives no longer describes what is happening. Apply the distinction to your own monthly deck before you change anything else: if it cannot separate the two columns, it will misread the next two quarters.

Why does a page rank at position 3 and earn zero clicks?

---

Because a high ranking now frequently sits underneath a generated answer that resolves the question without a visit. In our 90-day window, the query “enterprise seo workflows automation” put one of our pages at position 3.8, produced 1,618 impressions, and returned zero clicks (Indexable, 2026). That is not a title-tag problem. The query was answered above the result.

This was not one unlucky page. In a separate 28-day scan of the same property, at least 543 page-one queries returned zero clicks against 7,038 impressions (Indexable, 2026). Eight queries earned any click at all, and all eight were brand or navigational. The pattern is consistent with the wider market: SparkToro and Similarweb put US zero-click searches at 68.01% for January to April 2026 (SparkToro, 2026).

The operational consequence is that impressions and clicks now measure different things. A rise in impressions with flat clicks is not automatically a failure. It may be a page being seen and consumed, with the visit landing elsewhere or nowhere. You should decompose the two before assigning a cause, because the remedies are opposite: a genuine CTR problem wants a better title, and a zero-click surface change wants a different metric.

Do the pages that rank actually get cited?

---

Often not, and this is the finding worth taking to a board. We joined 223 of our own Search Console pages to the URLs that AI engines actually cited (Indexable, 2026). Eighteen pages sat in Google's top 10. Three of those were cited by an AI engine. Fifteen ranked and were never retrieved — 83%.

The inverse held too. Our two most-cited pages ranked at positions 30.5 and 28.1, which is page three of Google (Indexable, 2026). The engines were retrieving material Google's ranking did not favour, and ignoring material it did.

Who gets cited instead is mostly publishers and communities. Across the engines we track, the most-cited domains were semrush.com with 85 citing responses, youtube.com with 83, ahrefs.com with 62, indexableai.com with 52 and reddit.com with 52 (Indexable, 2026). Two of the top five are user-generated platforms no marketing team controls. If a third of the citation surface in your category sits on YouTube and Reddit, a content plan consisting only of pages on your own domain is addressing part of the problem.

Can you plan around AI Overviews?

---

Not reliably, because coverage is uneven in ways that do not follow intent. We checked nine enterprise-modified search terms live on 9 September 2026. Five returned an AI Overview; four did not (Indexable, 2026). Testing four matched pairs, adding the word “enterprise” removed the AI Overview on three of them — “seo agency”, “seo analytics” and “content marketing” each carry one while their enterprise variants do not.

That is a small, hand-picked sample in one country on one day, and AI Overview presence is volatile and personalised. The honest reading is not that enterprise terms are protected. It is that the same buyer, researching the same category across one morning, meets a generated answer on some queries and ten blue links on others. Any plan assuming uniform AI Overview coverage across an enterprise term set is assuming something we could not reproduce.

One methodological warning, because it cost us a false claim before we caught it. Keyword tools and live SERP fetches disagree. Ahrefs Keywords Explorer reported no AI Overview for “enterprise technical seo” on the same day its own live SERP fetch returned one at position 1 (Indexable, 2026). You should source presence claims from a live SERP check with a positive control, never from a cached features field. We ran the control first; without it we would have published a suppression pattern that does not exist.

What decision does the executive now own?

---

Three, and none of them are keyword decisions.

What you are optimising for on each surface. Ranking and retrieval need different work. A page built to rank is comprehensive; a page built to be retrieved is extractable — self-contained claims, question-shaped headings, visible dates. Deciding which pages do which is an allocation decision, and it belongs to whoever owns the budget.

What counts as evidence. If reporting is clicks-only, a programme growing its citation share looks like a programme in decline. Reporting has to carry both, and the executive should agree in advance that a click drop alongside a citation rise is not a failure. Get that agreement in writing before the next quarter closes, not after.

How long you wait. Citations do not appear on publish. Profound's 2026 analysis of ChatGPT and Claude agent logs across roughly 900 pages put the median lag at 6.81 days and the 90th percentile at 37.10 days (Profound, 2026). A 30-day review window will misread a meaningful share of the work as having failed. Schedule the review at 45 days instead.

What are the anti-patterns?

---

Five recur, and each has a specific correction you can apply this quarter.

Reporting clicks as the whole picture. When impressions rise and clicks fall, decompose the two before assigning a cause. The answer is often surface change, not page quality.

Optimising the head term and ignoring the question. On our domain the head terms sit between positions 76.8 and 89.0 while full-sentence questions sit between 1.9 and 5.3 (Indexable, 2026). Demand arrived in a shape the keyword plan had no row for. Use this as a prompt to re-cut your keyword research into questions.

Treating length as depth. The citation window sits between 500 and 2,000 words (AirOps, 2026). Beyond roughly 3,000 words a page surfaces a shrinking fraction of itself to retrieval systems. Long is not thorough.

Assuming schema is a ranking lever. It is a retrieval lever, and a modest one — the measured lift from JSON-LD is 6.5 percentage points, with tables and lists adding 2.9 (AirOps, 2026). Implement it as hygiene, then stop talking about it.

Judging the programme at 30 days. Given a 90th-percentile citation lag of 37.10 days (Profound, 2026), a 30-day verdict is a coin toss dressed as a decision.

How do you score your own programme?

---

Run this before you commission anything. Each answer is yes or no, and the whole thing takes an afternoon.

1. Step 1 — build the disagreement list. Export your top-100 pages from Search Console by impressions, then pull the URLs cited by AI engines in your category. Join them. Any page in the top 10 with no citation is your gap list.
2. Step 2 — split the report. Separate impressions and clicks into two lines with two owners and two explanations. Do this before the next monthly review.
3. Step 3 — re-cut research into questions. Take your top 20 head terms and write the full-sentence question a buyer would actually type. Check where you rank for the question, not the term.
4. Step 4 — verify AI Overview presence live. For those same 20 terms, run a live SERP check with a positive control. Do not trust a cached features field.
5. Step 5 — name the domains cited instead of you. If YouTube, Reddit or a review platform dominates, that is a channel decision, not a content decision.
6. Step 6 — set the evaluation window at 45 days and get sign-off that a click decline alongside a citation rise is an acceptable result.

Fewer than four yeses across steps 1 to 6 means the programme is being managed on ranking alone. That was sufficient until recently. On the evidence above, it no longer describes the whole outcome.

In summary

---

Enterprise SEO did not get harder. It got a second scoreboard, and most organisations are still reading the first one. The next step is small and cheap: build the disagreement list in step 1, and see how many of your ranking pages are invisible to the systems now answering your buyers' questions. On our estate the answer was fifteen out of eighteen.