Last updated: 7 September 2026

The AI Visibility Baseline: 700 Real Scans (2026 Data)

Across 700 live AI engine scans of four UK brands' target keywords, run between July and August 2026, the brands were named in the answer just 3.3% of the time. This article publishes the full method, the maturity breakdown behind that number, and why being cited as a source arrives long before being recommended by name.

Key takeaways

• Across 700 live AI engine scans of four UK brands' target keywords (July–August 2026), the brands were named in the answer in 23 scans — 3.3% overall.
• Brand maturity is the strongest signal in the dataset: the security services brand, with 600+ published articles and 15+ years trading, was named in 6.3% of its 270 scans; two brands in their first week of publishing scored 0% across 160 combined scans.
• The scanned engine read the brands' own websites far more often than it named the brands in text — being cited as a source consistently arrives weeks or months before being recommended by name.
• Across 863 scored, published articles from the same brands in the same period, the average quality score was 81.2/100 — publishing quality alone did not shorten the authority timeline.

What is an AI visibility baseline?

An AI visibility baseline is a measured starting point — not a target — that records how often a real AI engine names a specific brand in response to that brand's own buying-intent keywords, before any optimisation work begins. This dataset is one such baseline: 700 scans across four UK brands (a security services company, a marketing agency, a B2B software product and an AI software product) run between July and August 2026 using Aether AI's scan infrastructure.

Unlike a target — "we'll get you cited in 40% of answers" — a baseline is falsifiable: it can be re-run on the same question set and checked against the original figure. That distinction matters because most AI-visibility marketing consists of targets with no published method, which is exactly the gap this report is built to close.

Honesty notes, in line with the site's methodology: these scans ran on one engine — Anthropic's Claude, with live web search enabled — during this window, because API (application programming interface, the interface other tools use to query an engine programmatically) access to additional engines was constrained at the time. A further 83 rows from demo environments were excluded from every figure in this report. Every number here states what was counted out of what.

Why a baseline matters more than a target

A baseline is a number a reader can re-run; a target is a promise with no method attached. Any provider quoting a high day-one AI-visibility figure without disclosing scan volume, engine and date range is describing a target, not a baseline. This report exists to give growing brands a re-runnable reference point instead.

The baseline: 3.3% of answers name the brand

Of 700 scans, the target brand was named in the answer 23 times — a rate of 3.3%. These were not vanity queries built to flatter the result; they were the actual buying-intent keywords each brand targets, meaning competitive answers where the engine had many established names available to choose from instead.

This is the realistic starting line for a growing UK brand in 2026. Any provider quoting dramatically higher day-one figures should be asked to show the underlying scan data — how many scans, which engine, over what date range — because without that disclosure the number cannot be checked. For brands wanting to check their own starting point before committing to any programme, the free AI visibility audit runs the same style of scan against a brand's actual market.

Maturity is the strongest signal in the data

Brand age and publishing history, not article quality alone, explain most of the variation in this dataset. The security services brand — 600+ articles published, 15+ years trading — was named in 6.3% of its 270 scans, nearly double the overall average, while both brands in their first week of publishing scored 0% across a combined 160 scans.

This pattern holds across everything Aether AI measures elsewhere: engines learn sources before they learn brands. A new brand's articles can be read, retrieved and even cited by an engine while the brand name itself still goes unmentioned in the answer text — which is why the next section separates the two.

The gap between cited and recommended

Being cited means the engine read a brand's page and used it to build an answer; being recommended means the engine named that brand as the choice in the answer text. The security services brand's 6.3% mention rate sat well below how often its pages were actually retrieved as sources across the same 270 scans, confirming citation arrives first.

Being cited is a content problem and typically resolves within weeks of consistent publishing. Being recommended is an authority problem — entity signals, third-party evidence, elapsed time — and resolves over a longer horizon. A programme that measures only one of these two produces either false comfort (high citation, low naming, read as failure) or false despair (early naming mistaken for citation, and vice versa). Readers building their own tracking process can see how a structured checker separates these two metrics in the AI Visibility Checker guide.

What the losing answers cited instead

When the four brands were absent from an answer, the engine overwhelmingly cited independent blogs, review platforms and category round-ups rather than any vendor's own website. That pattern points to an off-site evidence layer that on-page publishing alone cannot close — a brand's own articles compete with, but do not replace, third-party coverage of that brand. The practical response to this gap is covered in how to appear in Google AI Overviews.

What doesn't shortcut the authority timeline

Publishing quality alone does not shorten the time it takes an engine to start naming a brand. Across 863 scored, published articles from these four brands in the same July–August 2026 window, the average quality score was 81.2/100 on Aether AI's internal quality gate — structurally sound, sourced, machine-readable content — and the overall mention baseline still sat at 3.3%.

Quality is necessary but not sufficient: unreadable or unverifiable content is never cited at all, so the quality floor matters. But once content clears that floor, further quality gains do not compress the authority timeline the way more off-site evidence and elapsed time do. Brands evaluating any AI-visibility software against this reality should read AI SEO Software UK 2026: What It Is & How to Choose before assuming publishing volume alone will move the number.

What this means if you're starting now

1. Set baseline expectations honestly. If a vendor promises immediate AI recommendations with no published scan data, ask to see the method: engine, volume, date range.
2. Measure from day one. A 0% baseline, like the two first-week brands in this dataset, is not failure — it is the number progress gets judged against later.
3. Watch citations before mentions. A site appearing as a source in AI answers is the leading indicator that being named by the engine comes next.
4. Invest where engines look for proof — reviews, communities, press coverage, public records — not only on owned pages.
5. Re-run the same question panel monthly. A fixed question set, repeated on a schedule, is the only way to separate real movement from ordinary engine drift.

Your AI visibility baseline checklist

• Fix your engine, question set and date range before the first scan — never change the panel to chase a better number.
• Run at least 30-80 buyer-intent questions per brand; below roughly 30, engine-to-engine variation swamps the signal.
• Record both the citation rate (source appears) and the mention rate (brand named) as separate figures every time.
• Exclude demo, test or non-production environments from every reported figure, and state that exclusion.
• Re-run the identical panel monthly for at least 12 weeks before drawing a conclusion about trend.
• Publish the method alongside the number — engine used, scan count, date range — so the figure can be checked.
• Keep the hardest, most competitive questions in the panel permanently; do not swap them for brand-name queries you already win.

Frequently asked questions

Is a 3.3% mention rate bad?

No — for a young brand in a competitive category, 3.3% is a normal starting point rather than a poor result. What matters is the trend across a fixed question set measured over months, not the opening number in isolation.

How many prompts do you need for a reliable baseline?

Aether AI's panels run 40-80 buyer questions per brand, because below roughly 30 questions, engine-to-engine variation swamps the signal and makes month-to-month comparison unreliable. The full approach is described on the methodology page.

Do these numbers apply to every industry?

No — categories with heavy independent review coverage, such as software and consumer goods, behave differently from service categories where engines lean more on directories and local listings. Treat this dataset as a reference point for UK B2B and services brands specifically, not a universal figure.

How often should a brand re-measure?

Monthly, at minimum. AI answers drift week to week as engine indexes refresh and competitors publish new content, so a single snapshot proves very little — every figure in this report comes from repeated scans rather than one isolated run.

Can I see where my own brand stands?

Yes — the free AI visibility audit runs this style of scan against a brand's actual market and reports what the engines say today, using the same method described above.

How we calculate the AI visibility baseline

Every figure in this report comes from the same repeatable process, described in full here so the AI visibility benchmark can be checked rather than taken on trust.

The scan itself

Each scan puts one keyword to a live engine, with web search enabled, phrased exactly as a buyer would ask it. Aether AI's scan process records the full answer text, every URL the engine cited, whether the brand name appears in that text, and which competitor names appear alongside it. Nothing is sampled or smoothed: a scan either named the brand or it did not, with no partial credit.

How the AI citation rate is derived

The AI citation rate counts answers where the brand's own domain appears among the engine's cited sources, divided by total scans. The brand mention rate counts answers that name the brand in the visible answer text, divided by total scans. Keeping these two figures separate is deliberate: a brand can be read constantly by the engine and named in the answer never, which is exactly the pattern this baseline data exposes in the maturity table above.

Why the brand mention rate moves over time

A brand mention rate figure taken once, on a single date, is close to meaningless, because engine indexes refresh, competitors publish new content, and underlying models update between scans. Taken monthly against a fixed question set, the same figure becomes a defensible measure of progress — which is why this report states its date range (July–August 2026) rather than presenting the number as a permanent constant. Teams comparing GEO (Generative Engine Optimisation — the practice of improving how brands appear in AI-generated answers) platforms on this basis should see GEO Platform Guide 2026: What It Is & How to Choose for how different tools structure this kind of tracking.

What we are changing based on this data

Three changes, stated now so they can be judged against the next baseline:

• More off-site work, earlier. Because the losing answers cite third-party sources rather than vendor pages, on-page publishing alone cannot close the gap identified above.
• Cluster depth over topic breadth. Within this dataset, brands were named most often where several related articles existed on a topic, not where a single strong article stood alone.
• Longer measurement windows. Twelve weeks is the shortest window in which engine drift can be reliably separated from real movement in the mention rate.

The next baseline will be published whether it improves on 3.3% or not.

How this squares with what the engine makers say

Google's Search Central documentation on AI features states there is no special markup or setting that gets a page into AI Overviews: eligibility follows ordinary crawlability, indexing and content-quality signals, the same signals that govern traditional search. Anthropic's documentation on its web search tool describes Claude retrieving and citing sources at answer time rather than drawing from any curated vendor list, and OpenAI's web search tool documentation describes the same retrieve-then-cite pattern for its own models.

Read together, these three sources explain the 3.3% baseline better than any theory of "AI SEO" does. If engines assemble answers from whatever the open web makes easy to retrieve and safe to quote, a young brand's problem is not a failed hidden optimisation test — it is that not enough of the open web, including that brand's own pages, yet says anything specific, sourced and current about it. Brands wanting a structured way to close that gap can compare options in the LLM SEO Tool: The 2026 UK Buyer's Guide.

That is a slower problem to fix than a meta tag, and a more durable advantage once fixed. It is also why this report tracks the citation rate and the mention rate as two separate lines: the first responds to publishing within weeks, and the second follows only once the wider web catches up.

The measurement trap to avoid

The temptation, when a baseline reads low, is to swap in a friendlier question set built from brand-name queries the brand already wins. Doing so produces a chart that climbs while the underlying business result does not move. The losing questions stay in this panel permanently for that reason, and any brand building its own baseline should apply the same rule.

Where Aether AI fits this baseline

This report exists because Aether AI runs the same style of citation scan for its own visibility tracking that it makes available through its platform — the scans behind this article's 3.3% figure were produced using Aether AI's scan infrastructure, not a third-party audit tool. That is the core proof point behind the platform: the citation tracking, keyword tracking and quality scoring described in this report are the same features available to any brand running its own baseline.

Aether AI tracks citations across six AI engines — ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini and Copilot — alongside keyword and competitor tracking and Google Search Console integration, with public pricing and no requirement to speak to a salesperson first. Brands wanting to see their own starting number before committing to a monthly programme can run the free AI-visibility audit, or review how the platform's approach compares to other tools in the GEO Tool Guide 2026.

Conclusion

This baseline is published because the alternative — quoting flattering numbers with no method attached — is how this category earned its credibility problem. A 3.3% mention rate is not a good headline. It is, however, a real one, measured against 700 disclosed scans, and it is the number this dataset's own future work now has to beat in public.

The next baseline will be republished as the dataset grows and engine coverage widens. The schema, method and prompt sets behind it remain public.