ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity and Claude. Measured one by one.
Multi-Platform Coverage for AI Search
An answer engine optimization tool is only as useful as the engines it asks, and how it asks them. 99Visibility audits ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity and Claude, scores each one separately, and tells you which answers searched the web and which came from memory.
The engines do not work the same way. Some search the web for each question and cite what they read; ChatGPT and Claude decide per question whether to search, the way their apps do, so one column can hold both kinds of answer. A page you fix today can change an answer that searched, and cannot change one given from memory until the vendor ships a newer model.
Google alone runs more than one: it says AI Overviews and AI Mode "may use different models and techniques" (Google Search Central, AI features guide). We read AI Overviews and AI Mode from Google's results, query Gemini through its API, and report all three as separate columns.
Read the columns before the headline. Each one shows how one engine treats you, and whether an answer searched or came from memory tells you which kind of fix can work.
ChatGPT
36% of market
9 of 10 answers name you
Web search on
Google AI Overviews
18% of market
0 of 10 answers name you
Web search on
Google AI Mode
10% of market
0 of 10 answers name you
Web search on
Gemini
9% of market
0 of 10 answers name you
Web search on
Perplexity
18% of market
7 of 10 answers name you
Web search on
Claude
9% of market
5 of 10 answers name you
Web search on
Coverage: named on 3 of 6 platforms, weighted by market share× 0.70
What you get
A column per engine
Each platform gets its own score from the same four metrics, without the coverage multiplier. A weak column stays visible even when the overall number looks fine.
Searched or from memory, stated
Every engine in the audit can use the live web. ChatGPT and Claude decide per prompt whether to search, and each of their answers is marked "Searched the web" or "Answered from memory".
Citations from every engine
Links are counted on every engine, and every engine can cite when it searches. An answer given from memory links nothing and counts as 0, so read a low citation rate beside how the answers were given, not as a verdict on your site.
Weighted coverage
Coverage weighs each platform: ChatGPT 36%, Google AI Overviews 18%, Perplexity 18%, Google AI Mode 10%, Gemini 9% and Claude 9%. Being named on ChatGPT moves your score more than being named on Claude.
How it works
- 1
Every platform in your plan, at once
Each prompt goes to ChatGPT and Google AI Overviews on Free, and to all 6 platforms on paid plans.
- 2
Each answer stays in its column
An answer that fails or times out is left out rather than scored zero.
- 3
No overview is not a miss
If Google shows no AI Overview for a prompt, that prompt is skipped on that column. There was nothing to be in.
- 4
Coverage ties them together
The platforms that answered set the coverage multiplier, weighted as above. Mentioned on none, the overall score is 0.
What we ask, exactly
This table is generated from the configuration the audit runs on.
Example output
| Platform | How we ask it | Web search | Coverage weight | Training data ends |
|---|---|---|---|---|
| ChatGPT | OpenAI API, gpt-6-luna, with web search | On | 36% | May 2026 |
| Google AI Overviews | Read from Google's results page in your market | On | 18% | Not published |
| Google AI Mode | Read from Google's AI Mode in your market | On | 10% | Not published |
| Gemini | Gemini API, gemini-3.8-flash, grounded on Google Search | On | 9% | Not published |
| Perplexity | Perplexity API, perplexity/sonar, with web search | On | 18% | Not published |
| Claude | Anthropic API, claude-haiku-5-5, with web search | On | 9% | June 2026 |
We call each vendor's API, not the consumer app, so each API column is a close proxy for the app rather than a copy; Google AI Overviews and Google AI Mode are read from Google's own results. ChatGPT and Claude have web search on and decide per prompt whether to use it, as their apps do. If grounding fails on a Gemini call, that answer is taken without search.
From a real audit: one brand, three engines
On 15 October 2026 we asked ChatGPT, Perplexity and Claude the same five shopper prompts about sportswear, none naming a brand. ChatGPT (gpt-5.6-luna) and Claude (claude-sonnet-5) answered from memory with web search off; Perplexity (sonar) searched the web.
From a real audit
| Engine | How it answered | Brand mentions in its answers | Under Armour’s share | Links to underarmour.com |
|---|---|---|---|---|
| ChatGPT | From memory | 118 | 6% | None given |
| Claude | From memory | 42 | 12% | None given |
| Perplexity | Web search | 32 | 13% | 0 of 75 citations |
One brand, three pictures. ChatGPT wrote the longest lists and supplied 118 of the 192 brand mentions, so Under Armour’s pooled 8% share leans on the engine where it did worst.
Search is where a site can change the answer. In the shopper prompts none of Perplexity’s 75 citations went to underarmour.com, yet in answers to prompts that named each brand, Perplexity linked underarmour.com 14 times and puma.com 34 times.
For "best soccer cleats", Claude named Under Armour from memory and Perplexity, which searched, did not. The fix list reads that as an access question: the model knows the brand, and the live search did not reach a page that says so.
Models and settings as the product used them that day; it has since moved to the newer models in the table above, and ChatGPT and Claude now answer with web search on. Gemini's answers in that run were cut short by a token limit on our side and are left out, and Google AI Overviews was not part of it.
Read the AI visibility audit of Nike, Adidas, Puma and Under ArmourFix it with Agent V
Agent V is the assistant beside your dashboard. Ask it why one engine names you and another does not, and it reads that engine’s answers in your audit and your own page as text, robots directives included.
If your crawler access check shows a search crawler blocked, V drafts the exact robots.txt lines to let it in. You paste them; V never edits your site.
Anything that changes your account, like running an audit or adding a keyword, comes to you as a card to confirm. Nothing changes until you do, and V never edits your site. It runs on Anthropic's Claude models and is on every plan.
Who it is for
- Brands whose buyers use more than one assistant and need to know where they are missing.
- Teams deciding what to fix first: a block on an answer that searches the web can be fixed this week; a gap in one given from memory cannot.
What it does not do
- It does not cover Microsoft Copilot or Meta AI.
- It asks from one country at a time, your first market, in the language you write the prompt in. Gemini takes no location, so its answers are not localized.
- It does not average away the differences. Engines cite in different units and Google's own surfaces can disagree, so read the columns before the overall score.
- It cannot see everything inside a search. We get the answer text, the links an engine returns and, where the engine reports them, the queries it searched for.
- It does not use the paid tiers of the consumer apps. A paying subscriber may be served a different model than the column shows.
Questions
Why are Google AI Overviews and Gemini separate?
They are different products. Google says its AI surfaces "may use different models and techniques" and can return different links, so one Google column would hide which of them is wrong about you.
Why do some columns rarely cite my site?
An answer given from memory links nothing, and ChatGPT and Claude give some answers that way. Open the answer to see whether it searched; if it did and linked other sites, those are the pages shaping it.
Which platforms does the Free plan cover?
ChatGPT and Google AI Overviews. Paid plans add Google AI Mode, Gemini, Perplexity and Claude.
Can a new brand appear in answers given from memory?
Not if it was founded after the model's training data ends, which is May 2026 for ChatGPT and June 2026 for Claude. It can still appear in answers that search the web, if the engine reaches a page about it.
What is answer engine optimization?
Working on how AI answers describe and cite you, the way SEO works on search results. For its AI features, Google says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary", so the work starts with being crawlable, indexable and quotable. For engines answering from memory, nothing published after their training data ends can appear.
Is AEO different from SEO?
For Google, no: its guide says "optimizing for generative AI search is optimizing for the search experience, and thus still SEO". What differs is measurement, because an answer has no position 1 to 10, so a tool has to read every answer and record whether you were named, how and with what links.
Is answer engine optimization real?
The part with evidence behind it is subtractive: crawler permission, firewall reachability, index eligibility and snippet controls decide whether an engine that searches can use your page at all. The best-designed test of rewriting content for citation found most methods "largely ineffective" (C-SEO Bench, arXiv:2506.11097). Treat it as finding and removing what blocks you, not as a formula for being chosen.
What should I look for in an answer engine optimization tool?
How it asks each engine, more than how many engines it lists. Check whether each column calls an API or reads the consumer product, whether web search is on and each answer says if it searched, and whether the full answer behind every number is there to read. The table on this page states the first two for every column we audit.
Explore More Features
See What AI Sees
Start on the Free plan. No card needed.