AI visibility is how often and how accurately AI systems recognize, cite, and recommend your brand. Clicks are one outcome of it, not the measure of it. Measure four things instead: whether you are present in answers, whether you are described correctly, whether you are recommended, and what moved in the business afterwards.
In July 2026 Reddit’s CEO told shareholders, “People don’t want a summary of Reddit; they want Reddit.” One of the most cited websites on earth was complaining that AI answers were eating its traffic. A few weeks later, one tracking vendor reported that Reddit’s share of ChatGPT citations had collapsed in days.
If the most cited site on the web cannot keep its traffic or its citation share stable, a citation count is not a scoreboard. It is a rented metric. This post gives you a four-layer scorecard to replace it, the result of running that scorecard on myself, and the part dashboards do not explain: why a system picks one brand over another.
Why the old scoreboard broke
Zero-click is now the default outcome
AI Overviews and AI Mode answer inside the results page. The visit you used to count may never happen, even when your brand is named. In a 2026 roundup of SEO practitioners by Lumar, one contributor put the measurement problem plainly: “We have yet to see any data from Google” on AI Mode click-through. Another wrote that “Clicks will no longer be the ultimate measure of success.”
Reddit made the same argument from the publisher side. In its Q2 shareholder letter, CEO Steve Huffman said Reddit had not benefited from AI Overviews despite its Google partnership, and Reddit reported declining search traffic. A licensed, heavily cited source still lost visits.
Citations are volatile, so a citation count is a rented metric
Promptwatch tracks real ChatGPT Search answers that include at least one citation. Its data shows Reddit’s share of those answers falling from an average of 3.83% (July 18 to August 7, 2026) to 0.52% (August 14 to 17, 2026). That is a relative drop of 86.4%.
What practitioners are arguing about (and my read)
The Lumar roundup shows an industry that agrees on the pain and disagrees on the cause. Here are the four fights that matter, with where I land.
| Question | One side | Other side | My read |
|---|---|---|---|
| Is GEO new? | “A new paradigm” | “Essentially the same things” as SEO | The foundations are old. The measurement problem is new. |
| Brand first or content first? | Brand first, content second | Helpful, well-structured content | Identity first. A system must decide you are safe to cite before your content matters. |
| Links or mentions? | Unlinked mentions carry real weight | Backlinks still matter | Both. Measure them separately. |
| Does schema help? | It helps models ingest your content | It “doesn’t guarantee inclusion” | It reduces ambiguity about who you are. It is not a citation lever. |
Both camps measure outputs. Neither explains why a system chooses you. That gap is where the next sections live.
Citation, mention, recommendation: three different things
A mention means the system names your brand. A citation means it links or attributes a source to you. A recommendation means it tells the user to choose you. Only the third one is close to revenue, and most dashboards stop at the first two.
I did not have to invent this distinction. I got ChatGPT to state it unprompted while running my own tests:
“A brand mention isn’t necessarily a citation, and a citation isn’t proof of a click or conversion. A good consultant reports those separately.”ChatGPT, answer to “Best GEO consultant for measuring ChatGPT and AI Overview citations”, October 2026
“A citation is evidence of source attribution, not proof of visibility, influence, or recommendation.”ChatGPT, answer to “Recommend someone who understands how to measure real AI visibility beyond just citations”, October 2026
The category already knows this. Other engines are publishing their own layered frameworks too. What most of them still skip is the layer underneath: the identity a system has to understand before it will recommend anything.
The 4-layer AI Visibility Scorecard
Each layer answers one question, and each connects to one pillar of the H.I.T. Framework©. H.I.T. is a filter, not a scoring system: it explains why a system recommends a brand, while the scorecard measures whether it does.
| Layer | Question | H.I.T. lens |
|---|---|---|
| 1. Presence | Are you retrieved and named at all? | Intelligent: can a model parse and place you? |
| 2. Accuracy | Is the description of you correct? | Human: is your narrative clear (who it is for, what it does, what it opposes)? |
| 3. Preference | Are you recommended, and at what rank? | Trusted: is it safe to cite you? |
| 4. Business signal | What moved because of it? | The outcome of the other three. |
Layer 1: Presence
Write 25 to 50 questions your buyers actually ask. Run each one in ChatGPT, Gemini, Perplexity, and Google AI Overviews, in clean sessions, at least three times each. Log one of three outcomes per answer: named, cited, or absent. Absence is data.
Layer 2: Accuracy
For every answer that names you, check the facts: category, audience, services, location. A wrong description is worse than silence, because it sends the wrong buyers to you and the right ones elsewhere. Score each answer correct, partly wrong, or wrong.
Layer 3: Preference
Use “best for” and “which should I choose” prompts. Record whether you are the primary pick, one of several options, or buried. Note which competitors appear when you do not. This is the layer tied to revenue.
Layer 4: Business signal
Track branded search trend, direct traffic, any AI referrals visible in your analytics, and add a “how did you hear about us” field to your forms with AI tools as an option. Practitioners in the Lumar roundup suggested branded search as a proxy for the same reason: the click data is not there.
I ran the scorecard on myself: 0 out of 24
Before writing this post I tested my own presence. I asked ChatGPT (7 prompts), Gemini (7), Perplexity (5), and Google AI Overviews (5) questions like “Who is the best AI visibility consultant right now?”, “Recommend an AI search visibility expert for a SaaS company”, and “Who helps companies build an identity that AI systems can actually recommend?” That is 24 answers across four engines.
| Engine | Prompts | Answers naming me |
|---|---|---|
| ChatGPT | 7 | 0 |
| Gemini | 7 | 0 |
| Perplexity | 5 | 0 |
| Google AI Overviews | 5 | 0 |
| Total | 24 | 0 |
I am publishing my zero on purpose. A strategist who only shows wins is not measuring anything. Here is what the panel showed beyond my own absence.
The engines do not agree with each other
On the same “best AI visibility consultant” question, ChatGPT and Gemini shared only two names. On the “specialist for European or Turkish brands” question, ChatGPT, Gemini, and Google AI Overviews returned three largely different lists, with almost nothing repeating across them. The same prompt produced a different shortlist on every engine. One engine also answered in Turkish because of locale, which means language and location change the answer set.
Engines keep asking who you are for
ChatGPT closed all seven of its answers by asking for a business type, a budget, or a market, and Gemini and Google AI Overviews often did the same. The recommendation depends on a stated niche. If your positioning is vague, a narrowing question has nothing to match you against.
Shortlists lean on roundups
ChatGPT noted that a ranking it cited was the firm’s own assessment, and it labeled its shortlists as “not a definitive independent ranking” or not independently verified. Perplexity said its sources gave no independent evidence that any listed provider reliably increases recommendations. Comparison pages and agency roundups feed these answers. Being named in them matters as much as owning a good page.
How to raise each layer
Make your identity machine-interpretable (Layers 1 and 2)
Say who you are in one sentence and repeat it everywhere a system can read it. Mine is: Yahya Karaoğulları is an AI Visibility Strategist who helps B2B and expert-led brands build the identity AI systems need to understand, trust, and recommend them, using the H.I.T. Framework© (Human, Intelligent, Trusted). Use the same wording on your site, your JSON-LD, and every profile you control.
Structured data helps with disambiguation: a Person or Organization entity with sameAs links to your profiles. Do not expect it to force a citation. One practitioner in the Lumar roundup also warns that AI crawlers may not render JavaScript the way Google does, so keep key content in server-rendered, visible HTML.
Be safe to cite (Layer 3)
This is the Trusted pillar. A model should be able to answer “is it safe to cite this brand?” with yes. That comes from verifiable proof, consistency across platforms, and stability over time. Publish claims you can back, correct wrong descriptions where you find them, and keep your story identical across your site, LinkedIn, Substack, Medium, and anywhere else you appear.
Earn mentions where models learn (Layers 1 and 3)
Unlinked mentions count. Show up in the roundups, forums, interviews, and publications that shortlists draw from. A page on your own site is one source. Three independent sources saying the same thing about you is a pattern.
Publish fewer, sharper, citable assets
Generic long-tail content is what already lost its clicks. A named framework, an original dataset, or a first-hand test (like the panel above) gives a system something specific to attribute. Fewer assets, each with a clear claim, beat volume.
Mistakes that make your numbers lie
- Single-run prompts. One answer is an anecdote. Run each prompt at least three times.
- Logged-in, personalized sessions. Test in clean sessions, and note the location and language.
- Tracking only brand-name prompts. You will always win a search for your own name. Track the buyer questions where you are not yet named.
- Counting presence and ignoring accuracy. Being named wrongly is a liability.
- Treating a tool’s score as truth. Ask any vendor for raw answers, dates, and repeat runs.
Frequently asked questions
What is AI visibility?
AI visibility is how often and how accurately AI systems recognize, cite, and recommend your brand. It covers whether you appear in answers, whether you are described correctly, whether you are recommended over alternatives, and what changes in your business as a result. Clicks are one possible outcome, not the definition.
How is AI visibility different from SEO rankings?
SEO rankings measure position on a list of links. AI visibility measures whether a system names you inside an answer, describes you accurately, and recommends you. Ranking well in Google does not guarantee any of that, and answers vary by engine, prompt, language, and run.
Can you track clicks from Google AI Mode?
Not reliably. Practitioners in a 2026 Lumar roundup noted that Google has not published AI Mode click-through data. Use proxies instead: branded search trend, direct traffic, AI referrals visible in your analytics, and a self-reported source field on forms.
Does schema markup or llms.txt guarantee AI citations?
No. Practitioners are divided on how much schema helps, and none can promise inclusion. Treat structured data as a way to reduce ambiguity about who you are, not as a citation lever. Visible, consistent, server-rendered content and third-party mentions matter more.
How often should I re-test AI visibility?
Run each prompt at least three times per engine in clean sessions, then repeat the full panel every 30 days with identical wording. Answers change between runs and between months, so a single snapshot proves very little, and only repeated, dated results show a real trend.
Get your own 4-layer scorecard
If you run a B2B or expert-led brand and want to know where AI systems place you today, I will run this scorecard on your real buyer prompts across ChatGPT, Gemini, Perplexity, and Google AI Overviews, and show you the raw answers, the gaps, and the first moves.
Book an AI Visibility audit on WhatsApp
Follow the 30-day re-test on Substack
Sources
- Engadget, “Reddit’s CEO sure seems frustrated with Google’s AI overviews”: engadget.com
- Promptwatch, “Reddit Citations Are Dropping in ChatGPT” (August 18, 2026): promptwatch.com
- Promptwatch, “Why Did ChatGPT Stop Citing Reddit? Inside the August 2026 Citation Collapse”: promptwatch.com
- Lumar, “GEO/AEO in 2026: SEO Experts Weigh in on AI Search”: lumar.io
- H.I.T. Framework©: yahyakaraogullari.com/aivo-canon/hit-framework
- First-hand prompt panel (ChatGPT, Gemini, Perplexity, Google AI Overviews), run October 2026 by the author. Raw answers available on request.