Contents 11
AI visibility is not a single “position in ChatGPT.” We measure it as a set of separate signals: whether an AI answer names the brand for a non-branded buyer question, whether it links to the site, which page it selects, whether it describes the company accurately, and whether people actually visit from AI services. This article documents the method on UPLIFY’s own site, including our first dated baseline and its limitations.
Why an “AI rank” is not enough
A conventional search position can be tied to a query, country, device, and date. AI answers add another layer of variation. The system may rewrite the question, change its source set, or answer without links. Two people asking the same question can receive different wording and citations.
That is why “we rank first in ChatGPT” proves very little without a protocol. We need the exact query, date, platform, web-search mode, brand mention, selected URL, and evidence that the result can be reproduced.
We do not hide all of that behind one polished GEO Score. A working report needs at least seven measures:
- Mention share: the percentage of answers that name the
brand. - Citation share: the percentage that link to the site.
- Linked-citation share among mentions: whether users can reach the primary source.
- Cited-page coverage: which URLs are selected and whether language versions are confused.
- Competitor share: who is named alongside or instead of us.
- Entity accuracy: whether services, markets, specialisation, and limits are described correctly.
- On-site outcome: sessions, engagement, and conversions from AI referrals, provided tracking is valid.
Our SEO Score vs GEO Score guide covers the conceptual difference between those metrics. This article focuses on the operating protocol and UPLIFY’s first-party evidence.
Three measurement layers that should not be blended
1. Native search-platform data
Google has started rolling out a Generative AI performance report in Search Console. It shows pages and impressions in AI Overviews and AI Mode, with country, date, and device dimensions. This is the most reliable source for behaviour inside Google because no third-party tool has access to Google’s internal AI impression data.
The report is not yet available to every property. If it is absent, we do not replace those impressions with a third-party estimate. We retain conventional GSC data, record when the new report becomes available, and start a separate baseline from that date.
One caveat matters for August readbacks. Google reported an AI Search impression logging issue from 13 to 17 August 2026. That period should not be compared with an unaffected window without an explicit annotation.
2. AI referral visits in GA4
GA4 answers a different question. It does not tell us how often the brand appeared; it tells us which visits reached the site. OpenAI adds utm_source=chatgpt.com to ChatGPT Search referrals, so that traffic can be grouped separately.
Between 1 and 30 July 2026, UPLIFY OS recorded 44 sessions from chatgpt.com / ai-assistant. Those are real sessions, but they do not prove a particular prompt or citation. GA4 does not expose the prompt text, and some AI visits may arrive without a stable referrer. Referral traffic therefore complements the citation panel; it does not replace it.
We attribute leads to AI only after the conversion event has been validated. If tracking is not reliable, the report says N/A instead of inventing a channel result.
3. A controlled buyer-query panel
The third layer covers ChatGPT, Perplexity, Gemini, Claude, and other systems that do not provide site owners with complete impression data. We freeze a set of non-branded buyer questions, run them under the same conditions, and save the answer, named brands, cited domains, URLs, and date.
This is not an “official ChatGPT position.” It is a control sample that can show direction of change when the protocol remains stable.
The queries used for UPLIFY
The first baseline used six Ukrainian-language buyer questions:
яке GEO агентство обрати для просування бізнесу в Україніскільки коштує GEO просування сайту в Україніяк потрапити в рекомендації ChatGPT та Google AI Overviewsхто проводить аудит Facebook Ads для інтернет-магазину в Україніде замовити аудит Google Shopping та Merchant Centerяке performance marketing агентство обрати для інтернет-магазину
We deliberately excluded UPLIFY from the wording. A branded query tells us whether a system recognises the company. A non-branded query tests whether it selects the company while a potential client is still choosing a provider.
Each service panel should cover several intents: provider selection, scope, price, comparison, audit, and a concrete problem. We do not invent those phrases in isolation. In the 24 August SerpAPI snapshot, all three checked English queries showed an AI Overview: how to measure AI visibility, AI visibility tracking, and generative engine optimization measurement. Some results also contained People Also Ask, video, and ads. That mixed informational and commercial intent shaped this article: method, tools, limits, a worked example, and FAQ.
What the 30 July 2026 baseline found
We ran the six questions through four live API proxies with web search: OpenAI web search, Perplexity Sonar, grounded Gemini 2.5 Flash, and Claude web_search. All 24 probes completed successfully.
UPLIFY was mentioned with a link in 4 of 24 answers, a citation share of 16.7%. Two of the six queries produced at least one citation.
The result was highly uneven:
- Perplexity linked to UPLIFY for the Facebook Ads audit query;
- OpenAI, Gemini, and Claude cited the site for the
Google ShoppingandMerchant Centeraudit query; - the two GEO queries produced 0 citations from 8 possible probes;
- the broad performance marketing agency query produced 0 from 4.
The systems also selected different pages from the same cluster. For Merchant Center, they cited the audit page, a PMax/Shopping page, and even English or Russian variants. The problem was therefore not simply “low domain authority.” The cluster lacked a clear signal about which URL should answer a specific question.
That baseline defined the next work: strengthen the GEO service pages with direct buyer answers, clarify the roles of Merchant Center URLs, add verifiable evidence to the Meta audit page, and publish a dated first-party measurement methodology.
How mentions and citations are recorded
Every probe receives a small evidence packet:
- the exact frozen query text;
- language, market, platform, model or mode, and date;
- whether the request completed successfully;
- whether UPLIFY was named;
- whether a clickable
uplify.agencylink appeared; - which UPLIFY URLs were selected;
- recurring competitors and third-party sources;
- whether the description of UPLIFY was accurate;
- a screenshot or stored response payload where permitted.
We keep a separate change log: what changed on the site, why it changed, when it reached production, how it was checked, and when the next readback is due. Without that record, it is easy to credit the latest article or backlink for a change it did not cause.
UPLIFY OS connects these records to tasks, the content queue, the evidence journal, and monthly reporting. A client needs a reproducible conclusion, not a folder full of unlabelled screenshots.
What honest trend reporting looks like
The first measurement is a baseline, not a trend. We need at least one comparable rerun before claiming that AI visibility has increased. The full panel is repeated after 60–90 days. Critical queries may be checked more frequently as operational signals.
A comparison is valid only when the following remain aligned:
- query wording;
- language and target market;
- platform and web-search mode;
- number of repetitions;
- mention and citation counting rules;
- GSC and GA4 date windows.
New questions are not retroactively mixed into the old denominator. They become a separate cohort with their own start date. This lets us report the core panel moving from 4/24 to a new value while keeping an expanded panel separate.
The monthly report divides change into four parts: visibility, cited pages, AI referral traffic, and business outcomes. The first three may improve while the fourth remains N/A if conversions are not yet verified. That is not a failure, and it is not a reason to manufacture leads.
What happens after the measurement
We do not publish articles simply because “more content is needed.” The intervention depends on the gap revealed by the panel.
If a brand is absent from buyer queries, it needs clear service pages, third-party validation, and a consistent entity footprint. If the brand is named but not linked, we strengthen primary sources: methods, data, cases, accountable authorship, and passages that can stand alone as citations. If an AI system cites the wrong language page, we inspect canonicals, hreflang, internal links, and the role of each URL. If referrals arrive but do not produce outcomes, we inspect the landing page, intent match, and tracking.
For UPLIFY, the clearest gap was the GEO cluster. We therefore updated the English and Russian GEO service pages with direct answers about scope, provider selection, measurement, and the limits of guarantees. We also strengthened proof paths between audit and service pages. This article fills the next gap by publishing the method and a dated result that can be inspected and repeated.
What a client receives in the report
A useful AI visibility report answers five plain questions:
- Which buyer questions did we test?
- Where was the brand mentioned, and where was it linked?
- Which pages and competitors did the systems select?
- What changed against a comparable previous period?
- What work was completed, and when is the next readback?
The method, date, limits, and sources accompany the numbers. GSC SEO data, GA4 AI referrals, and controlled AI probes are shown separately. They are three lenses, not three values that can be added into one percentage.
For a starting benchmark, our SEO/GEO audit reviews technical access, page roles, entities, citability, competitive gaps, and measurement. Our GEO agency selection checklist is useful when comparing providers.
Frequently asked questions
Is there one official AI visibility metric?
No. Google can expose its own AI Search page and impression data in Search Console, but ChatGPT, Perplexity, Gemini, and Claude do not provide a universal site-owner metric. Native reports, AI referrals, and a controlled query panel answer different parts of the problem.
How often should AI visibility be checked?
Critical queries can be reviewed weekly as a signal. A complete baseline is better repeated every 60–90 days, or after a meaningful batch of changes, with the wording and conditions preserved.
Does a mention without a link count?
Yes, but it is a weaker result. A mention shows that the system recognises the brand. A citation provides a path to the primary source and a chance to earn referral traffic, so the two should be counted separately.
Does GA4 show the prompt that produced a visit?
Usually not. GA4 can show the referral source, landing page, and subsequent behaviour, but not the full prompt in an AI chat. Prompt-level analysis requires a separate controlled panel.
Can GEO work guarantee inclusion in ChatGPT answers?
No. An agency can improve access, structure, evidence, brand entities, and off-site signals, but it cannot control a third-party AI system’s source selection. A guarantee of a specific mention or position would be misleading.
Why not compare any two screenshots?
Because the result depends on the query, date, language, market, model, and search mode. Without a matching protocol, the difference may be random variation rather than the result of the work.
Method and limitations
The SERP research informing this article was run through SerpAPI on 24 August 2026 with separate Ukrainian, Russian, and English query sets. UPLIFY’s controlled GEO baseline is dated 30 July 2026. It contains 24 successful API probes and is not consumer-interface impression data for every platform. GA4 measures visits, not total appearances in AI answers.
The current 4/24 result should not be extrapolated to another query set or market. It becomes a trend only after a like-for-like rerun. We also do not connect visibility changes to leads until conversion tracking and attribution are validated.