Search results are turning into answers. Visibility is no longer decided by position alone, but by whether your content fed into the generated answer – a layer that can now at least be roughly measured.
Type a question into a search box today and you increasingly get a paragraph rather than a list: a summarised answer, three or four linked sources beside it, the familiar results below. The decisive question is no longer which place a page holds, but whether it is among the sources the answer was built from.
This is not academic. A Vienna tax advisory firm sitting fourth for a basic question used to have a calculable chance of a click. If a model answers it and the firm is missing from the source row, fourth place is worthless – the visibility has not got worse, it has moved. Describing that layer is the ambition of the AI Analytics section that the redesigned Semalt panel puts alongside the classic reports.
From ten blue links to a single answer
The classic results page was a menu: ten suggestions, a glance at title and description, a decision, a click. That act of choosing produced the visit, the click-through rate, and made optimisation measurable at all.
A generated answer removes that choice: the user gets a synthesis, often from several sources, with a narrow strip of citations many never expand. Behaviour changes in three directions at once.
- Fewer clicks at the same interest. Questions answered in two sentences rarely lead to a visit. The need is met before anyone opens the source.
- Longer, more conversational queries. People who know a machine is answering phrase things more fully: two keywords become a question with a place, a period and a condition.
- Shifted click quality. Whoever clicks after the answer wants to verify or commission work. Volume can fall while the value per visit rises.
The third movement is usually overlooked: a falling click-through rate is not automatically a loss, since the superficial visits may have gone while the serious ones remain. Anyone thinking purely in positions misses it – the position stays put, while a second stage has been built above it.
Position five and “cited as a source” are two different goals
It is tempting to lump the two together: good rankings surely mean a model draws on the page. Partly true. But the goals follow different logic, which is why a separate layer of analysis exists at all.
The contest for a place
A ranking is an ordered list: exactly one first place, and the fight for it is zero-sum.
- Measured per keyword
- One winner per place
- Observable day by day
Inclusion in a synthesis
One answer can name several sources. What counts is not rank but whether a passage settles the sub-question cleanly and verifiably.
- Measured per question
- Several sources side by side
- Observable only indirectly
A page can sit ninth and still be cited, because it holds the paragraph that resolves the question. A page in second place can be passed over because its text is marketing language: a model can do nothing with “tailored solutions for your success”, and a great deal with a dated threshold and its scope.
The second distinction is granularity. Rankings are measured per keyword; citations happen per statement. One well-structured guide page can appear in answers to twenty phrasings of a question without ever standing at the top. A keyword list is therefore too coarse for planning the AI layer.
The table below sharpens the contrast deliberately; in practice the two logics mix.
| Dimension | Classic SEO thinking | AI search logic |
|---|---|---|
| Headline metric | Average position, clicks, impressions, CTR | Share of answers citing the domain |
| Unit of measurement | A single keyword | A question or search intent |
| Target state | As high up the list as possible | Inclusion in a multi-source answer |
| Competitive picture | Domains ranking for the same keywords | Domains seen as credible in the topic space |
| Lever for improvement | Relevance signals, links, technical access | Clear facts, clean structure, dated statements |
| Feedback | Daily data from Search Console and rank tracking | Model-based estimate, blurrier and hard to standardise |
| Time to effect | Weeks to months, reasonably observable | Hard to schedule, tied to model updates |
The last row is the most uncomfortable. With rank tracking you see when something moves. With citation frequency you see it late, irregularly and without reliable attribution of cause. Anyone quiet about that is selling a precision that does not exist.
The AI competitiveness score and the Market Circle
The AI layer comprises six views. The first calculates a competitiveness score for your domain and shows the field around it in what the panel calls the Market Circle: competing domains sorted into three rings – top-tier, mid-tier and niche.
Each ring suggests a different response; treat all three alike and you plan too big or too small.
Top-tier – the yardstick, not the target
Domains with the greatest topical authority: portals, trade publications, national providers.
- Not a target. For a ten-person business in Vienna these are not competitors you catch. As a benchmark the ring gives frustration, not insight.
- A content map. It shows which topics a model treats as central in your field, and how deeply they are worked out.
- Copy structure, not volume. What matters is how these sites break questions apart, not their article count.
Mid-tier – where visibility is genuinely contested
Providers of comparable size, often from the same region or the same segment.
- The honest mirror. A Vienna plumbing firm finds businesses with a website of similar vintage and a similar service promise.
- Compare coverage, not design. What counts is which questions these domains answer and which they omit. The omissions are your opening.
- Check the overlap. They are often not the domains from your rank-tracking list. That difference is the information.
Niche – where specialisation pays off quickly
Specialised domains with a narrow cut and few but deep pages.
- The most instructive ring for small firms. A site explaining one single procedure can turn up regularly in answers about it.
- Depth beats breadth. A generalist with twenty times the pages goes unmentioned, because none answers the question conclusively.
- Directly transferable. The same cut fits two or three of your own topics without rebuilding the site.
So: top-tier sets the topics, mid-tier sets the yardstick, niche supplies the tactics. Putting Semalt's AI Analytics views beside the classic competitor list from the SERP section shows the overlaps directly.
Market context: your own domain seen from outside
A second view generates a model-written market context for a domain you enter: positioning, traffic estimate, opportunities. It is prose plus figures, not measured analytics truth – and best treated that way.
Its value is the outside perspective. Companies almost always describe themselves differently from how their site reads to a model. A typical case: an agency sees itself as a shop-migration specialist, while four fifths of the site talks about web design – so the market context describes a web design provider. That gap is the information.
- Positioning. How the model files the domain thematically – compare it sentence by sentence with your own self-description.
- Traffic estimate. An order of magnitude for comparison, not a measurement. For small .at domains it can be well off.
- Opportunities. Starting points named by the model. Usable as ideas, unsuitable as a work plan without checking.
Query research: why intent overtakes search volume
The third view covers AI-assisted query research with a classification of search intent – a standard-sounding feature that shifts the weighting considerably in an AI context. In classic keyword thinking volume was the guiding figure: sort by volume, weigh difficulty against it, work the list.
With generated answers that has two weaknesses. First, long conversational questions are individually low in volume: a hundred variants at twenty searches a month appear in no volume-sorted list, yet together cover a topic area. Second, volume says nothing about whether a question leads to business. Intent classification sorts queries by what the searcher wants.
Knowledge questions
“How does X work”, “what does Y mean”. Many generated answers, little chance of a click – but the typical place where citation happens.
- Core terrain of AI visibility
- Needs closed answers
Comparison questions
“X or Y”, “what does X cost”. Users move on to a source more often here, because they want to secure a decision.
- Best mix of citation and click
- Needs figures and conditions
Action intent
“Hire X in Vienna”, “appointment in the 7th district”. Still mostly answered by classic results and local listings.
- Terrain of classic ranking
- Local signals stay decisive
Navigational intent
The search for a specific brand or page – least affected, because the destination is fixed.
- Barely any change
- No sensible lever to expand
For planning: AI visibility is decided on knowledge and comparison questions, while action intent stays the terrain of classic ranking. Both need content, built differently – and both can be planned separately once the query research inside the panel has sorted the raw list by intent.
Content gaps, leverage pages and the global visibility score
Two further views work on implementation: one analyses competitor strengths and content gaps, the other flags pages the model rates as leverage for expansion or internal linking. The gap analysis shows which topics the field covers and your stock lacks – less missing keywords than missing answers.
| Type of gap | How you spot it | First step | Effort |
|---|---|---|---|
| Missing cost answer | Mid-tier domains address prices, your site is silent | State a price range with conditions and date | low |
| Missing boundaries | Competitors explain when a service is not a fit | Put exclusion criteria in their own section | low |
| Missing procedural depth | Niche domains explain a process step by step | Add the process to the service page | medium |
| Scattered answer | The content exists, spread over four paragraphs | Question as a subheading, answer beneath | very low |
| Outdated figure | Numbers without a date, legal position years old | Add the date or update the value | very low |
The flagged leverage pages are the practical part: not new articles, but existing pages where expansion or better internal linking could have a disproportionate effect. For a small firm with forty pages, reworking three thoroughly beats ten shallow new ones. Tackle them alongside the technical groundwork – a page strong in substance but poorly reachable throws away both layers at once, as we set out under technical SEO.
The sixth view condenses a global visibility score for the AI search landscape: one value for a portfolio of domains, meant as a comparison over time. It compresses a great deal into one number and loses the causes. It can rise because new content lands – or because a competitor lost substance.
- As a trend line. The direction over three to six months tells you more than any single reading. A value with no history says nothing.
- As an anomaly alert. A sudden drop is reason to check the views underneath – not a finding itself.
- As a basis for discussion. A shared number saves endless case-by-case argument, provided you say what it is and is not.
What marks out content that gets used as a source
How do you write so a model finds the text usable? There is no secret formula, but six properties turn up in almost every analysis.
Clear facts
Numbers, deadlines, thresholds and conditions are usable. “Processing takes ten to fourteen working days” is quotable; “we work fast” is not.
Clean structure
One question, one heading, one answer. Spread it over four paragraphs and a subordinate clause and it is far less likely to read as a self-contained statement.
Unambiguous statements
Qualifications belong in the sentence, not the small print. Otherwise a model takes the claim without the condition – or not at all.
Currency
A visible date and a note on when the information applies help more than you would expect, especially on legal and tax topics with Austrian rules.
Verifiable figures
If you quote a number, say where it comes from and which period it covers. That is not just citation-friendly, it is sound work.
Clear authorship
A named author with a professional background, a proper imprint, a clear attribution. Anonymous content is weaker evidence.
The list matches what marks out good professional writing anyway. The difference is discipline: what once passed as stylistic sloppiness – the vague phrase, the undated number, the thought spread thin – is now a visibility disadvantage.
These limits argue not against measuring but for measuring correctly: as a compass for editorial decisions, not as proof of success. Proof stays where it was – with enquiries, closed deals and the hard data from the connected Google sources sitting inside the Semalt dashboard beside the AI layer. Having both in one interface is the practical gain: a hunch can be checked against click data at once.
Frequently asked questions
Does AI visibility replace classic ranking?
No. It adds a layer on top. Action-intent queries – someone in Vienna looking for a provider to hire – still run mostly through classic results and local listings. The shift affects knowledge and comparison questions. Neglect rankings while betting on AI visibility and you lose twice.
How accurate is an AI competitiveness score?
It is a model-based estimate, not a measurement. No official interface reports citations, so every tool approximates. The value is comparable over time within one system, not between providers. Use it for direction, not as a figure in a contract.
My company is small. Is the Market Circle worth looking at?
Yes, but with a different focus than a large provider takes. The outer ring is a map of the topic, the middle one a realistic yardstick, the inner one a tactical template. For small providers the niche ring is the most instructive: it shows how far a narrow specialisation carries without producing a hundred pages.
How long does it take for AI visibility to change?
There is no honest way to schedule it. With classic SEO measures, a first measurable movement typically shows after four to eight weeks. Inclusion in generated answers also depends on model updates nobody can plan around. Think in quarters and judge trends, not single readings.
Do I have to write entirely new content for the AI layer?
Only rarely. The biggest lever sits in what you already have: questions as subheadings, answers beneath them, numbers with a date and a scope, qualifications pulled into the same sentence. The flagged leverage pages are meant for exactly this – they show where a rework returns more than a new page.
A starting point you can manage in three months
For a company with limited resources, a short repeatable routine beats a large project – especially if each month ends with a tangible result, not another analysis.
- Month one – take stock. Look at the Market Circle, note the mid-tier group, compare the model-generated market context with your self-description. The discrepancy is the first assignment.
- Month two – gather and sort questions. Run the query research, group results by intent, pull out knowledge and comparison questions. Pick ten to fifteen that match the business you do.
- Month three – act where the leverage is. Rather than new pages, rework the flagged ones: questions as subheadings, answers beneath them, numbers with a date and a source, qualifications inside the sentence.
The shift from result list to generated answer does not devalue what came before; it lays a layer on top. Write cleanly structured, fact-solid content and you work on both layers at once – and measuring the new one roughly beats ignoring it.
To see the AI Analytics views for your own domain, you can set up an account in the Semalt panel and run the analysis on your own data. The score there is an orientation – but one you did not have before.