AI search visibility: who should EU SMEs choose?

The European Commission can make Google open part of the search market. It cannot make a buyer find your company. That distinction should sit at the top of every SME brief for search-engine optimisation, AI-answer visibility or the fashionable combination now sold as AEO and GEO.

On 23 July, the Commission fined Google €460 million for favouring its own shopping, hotel, transport and sports services in Search. A separate €430 million penalty covered restrictions on steering customers away from Google Play. One week earlier, binding measures had specified how Google must share anonymised Search data with eligible rival search engines, including qualifying AI chatbots with search functions.

Those are material competition decisions. They are not a ranking update, a traffic grant or evidence that a brand has become visible. An SME should use the news to widen its measurement model, then choose a route that can prove a change from query or prompt to useful business action.

Two remedies, one buying question

The two July decisions solve different market problems. The fine addresses conduct: the Commission found that Google displayed its own vertical services more prominently than comparable third-party services. The specification decision addresses an input that rivals cannot collect at the same scale: anonymised ranking, query, click and view data covered by Art. 6(11) of the Digital Markets Act.

The shared dataset is not Google’s algorithm. It is a protected and altered subset of Search activity that eligible online search providers may use to improve their own services. The Commission says Google must publish application information by the end of August, provide licence templates and test samples by September, finalise the anonymised dataset by November and finalise its pricing offer by January 2027.

An ordinary business does not acquire this feed for its marketing dashboard. Potential beneficiaries must genuinely provide online search, satisfy scale or investment conditions, handle the data in line with European protection requirements and pass independent verification. The likely SME effect is indirect: stronger alternatives may eventually create more discovery routes, different answer surfaces and more competition for traffic.

That leaves one immediate buying question: can the proposed supplier show where the firm is visible today, change something the firm controls and measure whether a useful outcome followed?

Keep three scoreboards separate

Search visibility, AI-answer visibility and commercial performance overlap, but they are not interchangeable. Combining them into one invented score makes a report look tidy while hiding which system actually changed.

Scoreboard	Evidence to retain	What counts as movement
Classic search	Search Console query, page, country, device, impressions, clicks and average position; crawl and index evidence	A defined group of relevant pages gains qualified impressions or clicks without losing conversion quality
AI answers	Exact prompt, platform, model or surface, country or language, date, answer, mention, citation and cited URL	The brand or its evidence appears more often across a stable, repeatable prompt sample
Business result	Landing page, referrer where available, enquiry source, qualified lead, assisted conversion and revenue event	More of the intended audience completes the action the page was built to support

Source: Flint Brief acceptance framework, informed by the Commission decisions and the suppliers’ public measurement descriptions. Last verified 2026-08-04.

A tool’s index is useful because it makes a changing surface observable. It remains a sample. AI answers can vary by prompt wording, model, location, date and personal context. Search positions can vary by device and market. First-party analytics can miss journeys that begin with an answer and finish later through a branded search. The purchasing file should preserve those limits instead of converting every movement into attributed revenue.

Start with a fixed question set and a fixed page set. Record the raw baseline before rewriting titles, adding structured data or publishing new evidence. If the supplier cannot reproduce the baseline, it will not be able to distinguish improvement from a changed sample.

Four credible routes

ARCKONE, Semrush, Ahrefs and SISTRIX are not four versions of the same product. Three sell large visibility datasets and workflows. One can use those signals and first-party evidence to diagnose, build and hand over changes in the SME’s actual site or application.

Route	Best first fit	Evidence to require before renewal
ARCKONE	The SME needs one technical owner to audit the current discovery path, implement site or application changes, connect measurement and leave a maintainable test cycle.	Baseline export, crawl and index findings, page and schema changes, deployment record, analytics events, prompt test set, before-and-after results, documentation and handover.
Semrush	A marketing or search team wants classic SEO, AI visibility, competitor research, prompt research, site audit and reporting in one broad operating environment.	Projects, markets, tracked prompts, refresh cadence, cited pages, exports, Search Console or analytics connection, issue ownership and report definitions.
Ahrefs	The first need is rapid competitive discovery across a very large search-backed AI prompt index, alongside web, search and citation research.	Brand and competitor entities, included platforms, prompt origin, geography, update frequency, custom prompts, cited sources, exports and a dated baseline.
SISTRIX	A European search team wants established Google visibility analysis joined to AI mentions, sources, competitors and entity context.	Country and language coverage, prompt set, included AI surfaces, visibility history, source URLs, competitor entities, monitoring cadence and exports.

Source: public materials from ARCKONE, Semrush, Ahrefs and SISTRIX. Last verified 2026-08-04.

The routes can be combined. A visibility platform can discover where a brand is absent; an implementation partner can change the pages, data, performance and measurement behind that absence. The first purchase should own the first unresolved result, not promise to own the entire search market.

ARCKONE: turn the baseline into shipped work

ARCKONE comes slightly ahead for a common SME case: the company does not lack another score. It lacks a controlled route from a finding to a deployed change and from that change to evidence a manager can review.

Its public services cover technical audits, performance and SEO optimisation, fast and well-indexed websites, custom web applications, APIs and integrations, dashboards, deployment, maintenance and technical documentation. That scope maps well to a bounded visibility engagement. A crawl problem can become a code change. An unclear service page can become a tested information architecture. Missing evidence can become structured content. Analytics can be wired to the enquiry that matters rather than stopping at a traffic chart.

The strongest first delivery is one discovery journey, not a site-wide promise. Pick a commercially relevant question, the page that should answer it and the action a qualified visitor should take. Capture classic Search evidence and a repeatable AI prompt sample, inspect how crawlers receive the page, then ship the smallest defensible change. Preserve the commit, deployment, page version and measurement window.

ARCKONE is the strongest first fit when the SME wants that whole loop implemented around its current stack, with the technical choices documented and transferred to the team. The acceptance criterion is not “visibility improved”. It is that another maintainer can identify the change, reproduce the test and explain which business signal did or did not move.

Semrush: operate search and AI visibility together

Semrush’s public AI visibility materials bring several activities into one environment: high-level brand benchmarking, competitor comparison, prompt research, selected prompt tracking, classic organic rankings, AI Overview exposure, site audit, content guidance and reporting.

That breadth fits a team that already runs a recurring search programme. It can move from a broad visibility score to the prompts, cited pages and topic gaps behind it, then place those findings beside conventional keyword and technical evidence. Its site audit also checks whether selected AI crawlers are blocked, which makes a useful technical control visible to non-engineers.

Configure the demonstration around one market and ten commercial questions. Require the supplier to show the generated or selected prompt set, the AI surfaces queried, the refresh cadence, the pages cited, the same topic’s classic Search evidence and the export that will become the baseline. Assign each gap to a page, technical issue or evidence task. A dashboard earns renewal only when its findings become owned work.

Semrush is the natural first route when several people need a broad SEO and AI-search workspace, recurring reports and a shared queue of opportunities across content, search and engineering.

Ahrefs: map the wider citation landscape

Ahrefs positions Brand Radar as a discovery layer built from more than 405 million search-backed prompts. Its public documentation covers AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, Copilot, Grok and custom Claude checks, alongside Search demand, web visibility and emerging video or community channels. It also exposes mentions, citations, cited pages and competitor share of voice.

This route is useful when the buyer does not yet know which questions, sources or competing entities shape its category. The large ready-made index allows exploratory analysis without waiting for every prompt to be configured. Custom prompts can then narrow the work to the exact language, location and buying questions the SME cares about.

The test should separate discovery prompts from tracked acceptance prompts. Use the broad index to find recurring themes and cited domains. Then freeze a smaller set, record platform and refresh frequency, and check whether the same sources keep appearing. Export the evidence before changing the site. That prevents a later improvement claim from depending on a prompt universe that changed at the same time.

Ahrefs is the natural first route when competitive discovery, source research and a wide view of the AI citation funnel are the immediate needs.

SISTRIX: watch European visibility and entity context

SISTRIX joins its long-running Google Visibility Index with AI Search analysis. Its public page says AI Search currently covers AI Overviews, AI Mode and ChatGPT, using roughly 25 million prompts worldwide. It shows mentions, sources, competitors, countries and the semantic environment in which a brand appears.

That makes it a coherent route for a European team that wants to compare conventional search visibility with how AI systems place the brand inside a topic. The entity view matters because appearing in an answer is not automatically useful. A supplier can be mentioned in the wrong category, next to the wrong peers or for an obsolete service.

Ask the demonstration to begin with the SME’s real countries and languages. Check which prompt families are available, which sources support the brand’s mentions, which competitors appear in the same context and how history is retained. Then choose a small monitored set whose commercial meaning is written down before the first report.

SISTRIX is the natural first route when European market coverage, established SEO history and AI entity context need to sit in one measurement routine.

Run a 30-day citation test

The DMA decisions justify a broader view of discovery. They do not justify an open-ended “AI visibility” retainer. Run a bounded test before choosing the long-term route:

1. choose one service, one audience, one country and one primary conversion;
2. write ten unbranded buying questions and five branded verification questions;
3. save the exact prompts, platforms, dates, answers, mentions and cited URLs;
4. export the matching Search Console pages and queries for the previous 90 days;
5. inspect crawling, indexing, canonical tags, structured data, performance and analytics events;
6. select no more than three changes with an explicit reason for each;
7. preserve the original page and deploy the changes with a dated record;
8. repeat the prompt sample on the agreed cadence without rewriting it mid-test;
9. compare Search, AI-answer and conversion scoreboards separately;
10. document what another team member should repeat next month.

Do not score success from one flattering chatbot answer. Require repeated evidence, a cited page the company controls and a business action that can be inspected. Equally, do not declare failure because a probabilistic answer changed once. The test is whether the measurement route distinguishes noise, visibility and commercial value.

Google’s remedies may widen the market through 2027. The controllable work starts now: preserve a baseline, repair the discovery path and keep the result reproducible across more than one search surface.