Above the familiar list of results a second shop window has opened: a written answer with a handful of sources beside it. A business absent there can rank perfectly well and still be invisible — and measurement of that layer has barely begun.
Until recently every SEO conversation opened the same way: where do we sit? The answer was a checkable number, and everyone knew what it was worth. That simplicity is gone. Above more and more queries now sits a composed answer with three or four sources beside it.
Take an Amsterdam advisory firm holding position five on a tax question for years. If the answer above it draws on three other domains, the position is unchanged and the return halves. Nothing dropped; something was added that intercepts the attention.
The AI Analytics section in the rebuilt Semalt panel tries to make that layer workable: six views estimating how a language model reads your market and where your domain sits in it. This article walks them through, caveat included — young tooling in a field with no agreed standard.
From a list of options to a composed answer
The classic results page was a menu: ten suggestions, and a user who picks. That act of picking turned visibility into a visit — it was what there was to optimise. A generated answer skips it: the summary is already there, and most readers never open the small print beside it.
- The easy questions vanish from your statistics. Anything settled in two sentences is settled inside the answer. Your page gets read without being visited.
- Queries grow longer and more spoken. Where someone once typed two words, there is now a full question with condition, deadline and location built in.
- The clicks that remain are worth more. Whoever clicks through despite a complete answer wants to verify, compare or make contact. Less traffic, higher average intent.
That third point is routinely misread. A falling click-through rate looks like loss on a chart, when it may equally mean the casual visitors dropped away and the serious ones stayed.
Holding position five or being cited: two different kinds of win
A good position does not produce a citation by itself. The link exists, but it is looser than expected and runs on different logic — hence a separate layer of analysis.
Taking a slot
Positions form a queue: one first place, taken by whoever displaces the previous holder. The contest is one of division.
- Unit: the keyword
- One domain per slot
Being admitted to the answer
One answer draws on several sources at once. What counts is not your rank but whether a passage settles a sub-question completely and checkably.
- Unit: the single claim
- Several domains side by side
This produces situations the old logic could not accommodate. A page at position nine gets cited because it alone names the exception the question was about. A page at position two is skipped because its text is promises: "we take the whole burden off your hands" gives nothing to quote; a deadline or an amount does.
The grain differs too. A position applies per keyword, a citation per claim. A well-built page can surface in the answers to fifteen phrasings of one question without ever entering the top three.
For Amsterdam businesses a second axis comes on top. A software company on the Zuidas, a boutique hotel in the Jordaan, a relocation agency for expats — each serves Dutch- and English-speaking searchers at once, usually from one domain. A country-level figure merges two markets with separate competitive fields and separate data densities.
| Aspect | Classic ranking | Answer layer |
|---|---|---|
| What you count | Position, impressions, clicks, click-through rate | Share of answers in which your domain appears as a source |
| Smallest unit | The keyword | The single, self-contained claim |
| Main lever | Relevance, links, technical health, speed | Factual detail, structure, a visible date |
| Where the measurement comes from | Search Console and ranking data, daily | Sampling and estimation, no official source |
| Response time | Weeks to months, reasonably trackable | Irregular, partly driven by model updates |
| Attributing causes | Workable, given enough data | Weak; effect and coincidence are hard to separate |
The bottom two rows are the awkward ones: you notice a change late, at uneven intervals, and rarely with a cause you can name.
The competition score and the three rings of the Market Circle
The first view calculates a competition score and arranges your surroundings in a figure called the Market Circle. Competitors sit in three rings — top tier, mid tier and niche — and each ring calls for a different decision.
The AI competition score
An estimated measure of the chance a model draws on your domain in this field.
- Comparable only within itself. Two scores from the same system can be laid side by side. A score from another tool means something else, even where the number matches.
- Readable as a movement. One reading says little. The direction over six months says something, provided the rest of your market held still.
- Unsuitable as a commitment. Never write this value into a contract or a bonus scheme. It can move without anything on your site having changed.
In practice the rings beat the number at the centre. They show who the model compares you with, and that list often differs sharply from the competitors you would name.
| Ring | Who sits there | What it gives you | Common mistake |
|---|---|---|---|
| Top tier | National portals, trade press, large providers with hundreds of pages | A map of the topics the model treats as the core of the field | Taking these domains as the target to match |
| Mid tier | Companies of your own order of size, usually from the same city | An honest benchmark: this is where visibility is genuinely divided | Skipping the ring because the names are unfamiliar |
| Niche | Narrow, specialised domains with few but deep pages | The best illustration of how far one thoroughly worked topic reaches | Dismissing them as too small to learn from |
For a fifteen-person firm in Amsterdam-Oost the outer ring is a table of contents, not a goal. The middle ring gives the yardstick, the inner ring the tactics: a small domain that writes out one working method in full recurs remarkably often. Hold the competitor views in the panel against the SERP section's competitor list too — they overlap only in part, and the domains missing from one deserve attention.
Market context: your domain as a model summarises it
The second view has the model write a market context: how it reads your positioning, an estimate of traffic, a list of openings. What comes back is text with a few numbers in it — description, not measurement.
Model-generated market context
The outside view: how your site reads to a system that has never met you.
- Positioning. Set the paragraph beside the wording you introduce yourself with and underline what is wrong. That short list is your first assignment.
- Traffic estimate. An order of magnitude. On smaller .nl domains the deviation runs high, because there is little reference material.
- Openings named. Prompts, not instructions. Test each suggestion against what you genuinely deliver, and to whom.
The value sits in the friction. An Amsterdam design studio sees itself as a specialist in brand identity for tech companies, yet gives three quarters of its site to one-off web projects — and the market context duly describes a web builder. That contradiction is the most useful sentence in the report.
Sorting keywords by intent rather than by volume
The third view runs AI-assisted keyword research and classifies what it finds by search intent. Under the old method volume was the compass: sort by volume, put difficulty beside it, work down from the top.
In the answer layer that fails twice. Written-out questions carry almost no volume each: eighty variants at fifteen searches a month sink to the bottom of any sorted list, though together they cover a whole area of demand. And volume predicts nothing about the work that follows.
Looking for an explanation
"How does X work", "what does Y actually involve". This is where most generated answers appear, and where a citation is earned.
- Heart of the answer layer
- Needs self-contained explanation
Weighing up and comparing
"X or Y", "roughly what does X cost". Readers click through more often here, because a decision needs something to rest on.
- Citation and click both
- Needs figures and conditions
Wanting to arrange something
"Notary Amsterdam Centrum", "meeting room hire Zuidas". These queries still run through the classic results and the map listings.
- Territory of the ranking
- Local signals stay decisive
Looking up a known name
Someone is after a specific brand or firm. The destination is fixed, so this is where least changes.
- Barely any shift
- No lever for growth
Visibility in the answer layer is decided almost entirely in the first two profiles; the third stays the domain of classic ranking. Those are two kinds of text, built differently, and they belong in separate plans once the keyword research Semalt produces has grouped the raw list by intent.
Content gaps: what your site does not say
Two of the six views are built for execution. One sets competitors' strengths against the holes in your content; the other flags existing pages the model treats as leverage for expansion or internal linking. The difference from classic gap research is fundamental: what is missing is not keywords but answers.
| Type of gap | How to spot it | First move | Effort |
|---|---|---|---|
| No indication of cost at all | Mid-ring domains name amounts or ranges, your site says nothing | Publish a range with its conditions and the date it applies from | low |
| No lead times | "How long does it take" questions are answered by other people | Give real durations per step, outliers included | low |
| No boundaries | Competitors explain when their service is not the right fit | Put the exclusion criteria under a heading of their own | medium |
| No English counterpart | In Dutch the question is covered, in English it stays open | Treat the question as a full text in its own right, not as a translation | high |
Those leverage pages are the most workable part of the AI section: not a wish list of articles to write, but existing pages where expansion pays off out of all proportion. On a fifty-page site, revising four thoroughly beats twelve new ones nobody finishes.
- Start with pages that already collect impressions. Where Google already registers interest, the threshold for being cited is lowest.
- Write the question as the subheading. One question, one heading, one answer underneath. Spread over four paragraphs, an answer stops being an answer.
- Put the number in the sentence itself. An amount in a footnote or an image does not count. What is not in the running text does not exist for this layer.
None of it survives a weak foundation: a page that loads slowly or is poorly reachable gets skipped by the answer layer too — more on that in the overview of what we do.
The global visibility value and what it leaves out
The sixth view condenses everything into one global visibility value within the AI search landscape: a single number for your whole portfolio, meant for comparison over time. Handy in a board meeting, dangerous once somebody treats it as a performance figure.
Global visibility value
One value for all domains together, intended to be read as a curve.
- As a curve, not as a standing. Only after three to six months does the direction carry meaning. A lone value with no history says next to nothing.
- As an anomaly alarm. A sudden drop is a reason to work through the underlying views — not yet a finding, and certainly not a conclusion.
- As a basis for discussion. One shared number saves endless argument over detail, as long as you say each time what it is not.
What gets lost is the why. The value can rise because your revised pages land, or because a competitor let their site thin out. On the chart both look identical.
What makes a text usable as a source
That leaves the practical question: how do you write so your text gets taken up? No formula exists, but certain properties recur in analyses of quoted passages.
Numbers instead of adjectives
"The procedure takes eight to twelve working days" is usable material. "We work quickly and carefully" is not, however true it may be.
Conditions inside the sentence
Limits belong in the same sentence as the claim. Park them in a separate disclaimer and the claim gets taken up without its condition, or not at all.
Visible currency
A stated effective date weighs more than expected, certainly on tax, legal and permit-related topics that differ by country and by year.
Traceable authorship
A named author, a findable office, a source under every figure. Anonymous text is awkward to use as evidence.
Much of this always applied to decent professional writing. What is new is the strictness of the settlement: what used to be taste now costs visibility.
A steady rhythm beats one large project. Month one: walk the Market Circle and compare the market context with your company profile. Month two: group the questions by intent and keep ten to fifteen. Month three: revise only the leverage pages. Then set the outcome beside the connected Google data in the same overview, so every assumption stands next to a measured click figure.
Frequently asked questions
Does the answer layer replace classic ranking?
No, it sits on top of it. Queries where somebody wants to arrange something — an appointment, a quote, an address — still run through the ordinary results and the map listings. The shift mainly affects explanation and comparison questions. Letting positions slide to bet on the answer layer loses on both sides.
How reliable is the AI competition score?
Treat it as a model's estimate, not a measurement. Since citations are officially published nowhere, every tool here approximates. Within one system the value is usable over time; between providers it is not. Use it for direction, never as a performance indicator.
We are a small firm. Is the Market Circle any use to us?
It is, only you read it differently from a national player. Take the outer ring as a map of topics, the middle ring as a realistic benchmark and the inner ring as an example. The last is the most useful for small firms: it shows how much reach one deeply worked specialism produces.
How long before something changes in this layer?
That cannot be dated seriously. With classic SEO work the first measurable movement usually arrives within four to eight weeks. Here, dependence on model updates comes on top, and that cannot be scheduled from outside. Count in quarters, and judge developments, not snapshots.
Our site serves Dutch and English-speaking clients. What does that mean?
That you get a different picture per language: different competitors, different volumes, a different density of data. In Amsterdam that is the rule, not the exception. Assess each language separately — a domain can be strong in Dutch and all but absent in English without a merged figure showing it. More background on our blog.
The composed answer does not make earlier work pointless; it adds a layer. Both layers ask the same things — clear structure, checkable facts, self-contained answers. What changes is the reporting. An estimate may be called an estimate; that is not weakness but a property of a field with no agreed standard.
To see what these views make of your own domain, you can set up an account in the Semalt panel and run the analysis on your data. The number stays an approximation — but an approximation of something you so far knew nothing about.
Hulp Nodig met Uw SEO?
Onze Amsterdam SEO-experts kunnen u helpen deze strategieën te implementeren en uw rankings te verbeteren.
Gratis Consult Aanvragen