Brand visibility in AI answers
Where is your brand in AI answers?
People now ask assistants the questions they used to type into Google. The answer names a handful of brands, cites a handful of sources, and nobody sees page two. LLMsMonitor measures where you stand inside that answer.
Built for the Turkish market: Hepsiburada'dan is a mention of Hepsiburada; HEPSİBURADA and HEPSIBURADA are one brand.
Same question · same day · Türkiye
The prompt asked:Which online marketplace is the most reliable for shopping in Türkiye?
In Türkiye, the choice of marketplace usually comes down to three things: how wide the catalogue is, how fast delivery is, and how smoothly returns are handled.
The platforms that stand out:
- Trendyol — the marketplace with the widest seller network; the catalogue is largest in fashion and home. Trendyol's own courier network shortens delivery times. () trendyol.com
- Hepsiburada — strong in electronics and white goods. A high share of items bought from Hepsiburada come from authorised sellers, and returns are handled from one screen. HEPSİBURADA Premium removes the delivery fee; outside campaign periods the price can sit above the marketplace average. () hepsiburada.com
- Amazon Türkiye — access to the international catalogue and a return policy that works the same across categories. On Amazon Türkiye the local catalogue is smaller than the marketplaces'. () amazon.com.tr
- n11 — offers a price advantage during campaign periods; category depth is mid-range. () n11.com
- Çiçeksepeti — specialised in flowers, gifts and food. () ciceksepeti.com
trendyol.com
hepsiburada.com
amazon.com.tr
n11.com
ciceksepeti.comHow to choose
- If warranty matters in electronics, Hepsiburada's share of authorised sellers is an advantage.
- If you follow campaigns, n11; if you are buying a gift, Çiçeksepeti is the better fit in its own field.
- For books, İdefix has the deepest catalogue in its category.
For anyone who needs same-day delivery, Amazon Türkiye's standard window runs longer than the marketplaces' in-city options.
One question, three assistants
Three assistants, three different answers
The same 12 questions, the same 30 days, the same market. They rarely agree with each other, and the disagreement is the useful part.
ChatGPT
47.3%
appeared in 88 of 186 runs
In the answer above: ranked 2
Gemini
34.1%
appeared in 62 of 182 runs
In the answer above: never mentioned
Google AI Overview
21.7%
appeared in 13 of 60 runs
In the answer above: ranked 4
38.1% overall — not the 34.4% that averaging the three rates would give. Why, below.
What is measured
Six numbers, all of them defined
How each one is calculated is written down; under every card, how many runs it rests on.
Visibility
38%
5.4% increase Compared with the previous period 5.4% increase
The share of answers that name your brand at all.
Based on 428 finished runs.
Sentiment
74%
3% increase Compared with the previous period 3% increase
How warmly the answers that do name you talk about you.
Based on 428 finished runs.
Mention density
1.4×
+0.2× Compared with the previous period +0.2×
How often your brand is named per answer.
Based on 428 finished runs.
Share of voice
21%
1.9% increase Compared with the previous period 1.9% increase
How much of the brand talk in your market belongs to you.
Based on 428 finished runs.
Average position
#2.8
0.4 decrease Compared with the previous period 0.4 decrease
Where your brand lands in the order brands get named.
Based on 428 finished runs.
Citations
34
11 increase Compared with the previous period 11 increase
How often your own pages are cited as the source behind an answer.
Based on 428 finished runs.
Citations
Publish where the assistant reads
The table customers actually pay for.
Most cited sources
The domains the assistants leaned on while answering, over the selected period.
| Domain | Citations | Answers | Share of all citations |
|---|---|---|---|
| sikayetvar.com | 47 | 31 | 22.6% |
| webrazzi.com | 38 | 26 | 18.3% |
| hepsiburada.com Your site | 34 | 25 | 16.3% |
| donanimhaber.com | 29 | 22 | 13.9% |
| eticaretmag.com | 26 | 19 | 12.5% |
| cimri.com | 20 | 15 | 9.6% |
| trendyol.com | 14 | 11 | 6.7% |
Something is read before an answer is written.
Which sites does the assistant cite when it talks about your market? In the example, sikayetvar.com was cited 47 times over 30 days, the brand's own site 34. This is not a ranking report; it is a publishing plan.
Two rules
How a number is made matters as much as the number
Two rules, in every query, every chart, every CSV.
Ratios are summed first and divided last.
| ChatGPT | 88 / 186 | 47.3% |
|---|---|---|
| Gemini | 62 / 182 | 34.1% |
| Google AI Overview | 13 / 60 | 21.7% |
| Average of the three rates | 34.4% | |
| Pooled 163 / 428 | 38.1% |
3.7 points apart
The denominators are not equal. If one assistant is read three times a week, its rate cannot weigh as much as the others. Whatever filter you look through, the arithmetic is the same.
A day with no runs is a gap, not a zero.
Three days not measured: a gap
The same three days counted as zero
The same data, two different claims. The lower chart reads a day nobody measured as "the brand never appeared". The distinction survives from the query into the chart, the sparkline, the table and the CSV.
Language
Turkish was not added later
We did not translate the interface; we built the counter for Turkish.
Assistants answer in Turkish, and Turkish inflects brand names. Hepsiburada'dan, Hepsiburada'nın, Hepsiburada'ya — all three are mentions of the same brand, and all three are counted.
Case folding is a separate job. HEPSİBURADA and HEPSIBURADA are one brand, but in regular expressions the lowercase of I is i, not ı, and İ folds to nothing at all. In Turkish both pairs are the other way round. So all four spellings of every term are written into the pattern by hand.
Counting is deterministic. Mentions and positions are counted by regular expressions, not by a language model. The model's job is sentiment, arbitrating ambiguous names, and suggesting brands you are not tracking yet.
| Seen in the answer | Counted? |
|---|---|
Hepsiburada'dan |
Mention of Hepsiburada |
Hepsiburada'nın |
Mention of Hepsiburada |
HEPSİBURADA |
Mention of Hepsiburada |
HEPSIBURADA |
Mention of Hepsiburada — with the dotless I |
Trendyol'un |
Mention of Trendyol |
hepsiburada.com (link target) |
Citation — not a mention |
Hepsiburada Seyahat |
Not counted — exclusion rule |
hepsiburada (in a source label) |
Not counted — citation labels are blanked before the scan |
Setup
Four short steps
The first answers start arriving within minutes.
-
1
Your brand
The brand name and domain we should look for in AI answers.
-
2
Market
AI answers change completely from country to country; pick the market and language you sell in.
-
3
Competitors
The brands you expect to see beside yours. As real answers arrive, we suggest new ones.
-
4
Prompts
The questions customers ask before choosing a brand like yours. Written the way a customer would type them, not as keywords.
Where the data comes from
Read from the product itself, never asked of an API
All four are read by driving the product: ChatGPT and Gemini through their real interfaces, the two Google surfaces from the Google results page. None of them calls a model's API with a question — because the answer an API gives is not the answer a person sees.
Pricing
As much as you measure
How many questions, how many assistants, how often.
Questions
Frequently asked
Is this an SEO tool?
No. SEO measures ranking; this measures the narrative inside the answer. The two can move in opposite directions on the same day — which is why we place the organic results for a query directly under its Google AI Overview answer. Seeing both side by side answers "who did the answer cite, and who did it rank" at a glance.
How many prompts should I track?
Ten to twenty is a good start. Write the questions a customer asks before choosing a brand like yours: "best", "which one", "compare", "would you recommend". Real sentences, not a keyword list.
If I ask the same question twice, do I get the same answer?
No, and that is exactly why the measurement rests on a period rather than one answer. Visibility is not "did it appear today" but "in what share of answers did it appear over the last thirty days". One answer is an anecdote; a hundred are a measurement.
How do you count Turkish suffixes?
"Koçtaş'tan" is a mention of Koçtaş, and so is "Koçtaştan". The inflected forms of every term are part of the pattern, apostrophe or not. Capital İ and dotless ı are handled explicitly, because regular expressions fold them wrongly.
Where is my data kept?
On our own server, in one database. No page on this site makes a request to a third-party server; every file, fonts included, comes from this domain.
Don't wait for the next answer.
Enter your brand, your competitors and twelve questions. The first answers start arriving within minutes.