AI recommendation index · August 2026
Which AI app builders do the AIs tell people to use?
When someone asks an AI how to build an app, Cursor comes up in about 1 of every 3 answers. Nothing else comes close.
Buyers ask AI assistants what to build with before they ever see a search result. Every month we ask ChatGPT, Claude, Perplexity and Grok the same 40 real buyer questions and count which tools they actually recommend. This is the scoreboard for August 2026.
- Questions
- 40
- Engines
- 4
- Answers analysed
- 160
- Run date
- 2026-08-20
By buyer intent · August 2026
Who is asking changes the answer
The same 40 questions, split by what the asker actually wants: to build without code, to work as a developer, or to get one specific thing built. Each board counts recommendations out of that segment's answers only, so a tool is judged against its own buyers rather than averaged across everyone else's.
Best for non-coders
When the asker says they cannot code, Lovable is the pick in 31 of the 88 answers.
22 questions · 88 answers
Best for developers
Ask as a developer and the answers change: Cursor leads with 23 of the 28 answers here.
- 1Cursor23 of 2882.1%
- 2GitHub Copilot12 of 2842.9%
- 3Claude Code11 of 2839.3%
- 4Windsurf6 of 2821.4%
- 5Aider5 of 2817.9%
7 questions · 28 answers
Best for a specific build
Name the thing you want built and Cursor comes up first, in 18 of the 44 answers.
- 1Cursor18 of 4440.9%
- 2Lovable16 of 4436.4%
- 3v012 of 4427.3%
- 4FlutterFlow10 of 4422.7%
- 5Bolt9 of 4420.5%
11 questions · 44 answers
All questions combined
The single table across every audience: how often each tool gets recommended, not merely named, when the AIs answer a buyer. Every number is a count you can check: "63 of 160" means the tool was the pick in that many of the 160 answers we collected. Per-engine columns count that engine's 40 answers.
| # | Tool | Share of voice | ChatGPT | Claude | Perplexity | Grok | Engines | Trend |
|---|---|---|---|---|---|---|---|---|
| 1 | Cursor | 63 of 16039.4% | 11/4027.5% | 15/4037.5% | 14/4035.0% | 23/4057.5% | 4/4 | first run |
| 2 | Lovabletry → | 47 of 16029.4% | 6/4015.0% | 11/4027.5% | 14/4035.0% | 16/4040.0% | 4/4 | first run |
| 3 | Bolt | 38 of 16023.8% | 6/4015.0% | 11/4027.5% | 8/4020.0% | 13/4032.5% | 4/4 | first run |
| 4 | Bubble | 35 of 16021.9% | 13/4032.5% | 6/4015.0% | 7/4017.5% | 9/4022.5% | 4/4 | first run |
| 5 | v0 | 32 of 16020.0% | 5/4012.5% | 7/4017.5% | 6/4015.0% | 14/4035.0% | 4/4 | first run |
| 6 | Replit | 31 of 16019.4% | 9/4022.5% | 9/4022.5% | 6/4015.0% | 7/4017.5% | 4/4 | first run |
| 7 | FlutterFlow | 31 of 16019.4% | 14/4035.0% | 7/4017.5% | 4/4010.0% | 6/4015.0% | 4/4 | first run |
| 8 | Glide | 26 of 16016.2% | 7/4017.5% | 6/4015.0% | 4/4010.0% | 9/4022.5% | 4/4 | first run |
| 9 | Softrtry → | 19 of 16011.9% | 6/4015.0% | 3/407.5% | 3/407.5% | 7/4017.5% | 4/4 | first run |
| 10 | GitHub Copilot | 18 of 16011.2% | 6/4015.0% | 5/4012.5% | 4/4010.0% | 3/407.5% | 4/4 | first run |
| 11 | Claude Code | 18 of 16011.2% | 3/407.5% | 7/4017.5% | 6/4015.0% | 2/405.0% | 4/4 | first run |
| 12 | Retool | 15 of 1609.4% | 8/4020.0% | 2/405.0% | 1/402.5% | 4/4010.0% | 4/4 | first run |
| 13 | Framer | 15 of 1609.4% | 5/4012.5% | 4/4010.0% | 0/400.0% | 6/4015.0% | 3/4 | first run |
| 14 | Adalo | 14 of 1608.8% | 2/405.0% | 4/4010.0% | 2/405.0% | 6/4015.0% | 4/4 | first run |
| 15 | Airtable | 12 of 1607.5% | 7/4017.5% | 2/405.0% | 1/402.5% | 2/405.0% | 4/4 | first run |
| 16 | Windsurf | 9 of 1605.6% | 2/405.0% | 2/405.0% | 2/405.0% | 3/407.5% | 4/4 | first run |
| 17 | Webflow | 9 of 1605.6% | 4/4010.0% | 1/402.5% | 1/402.5% | 3/407.5% | 4/4 | first run |
| 18 | Aider | 8 of 1605.0% | 0/400.0% | 2/405.0% | 0/400.0% | 6/4015.0% | 2/4 | first run |
| 19 | Continue | 6 of 1603.8% | 2/405.0% | 0/400.0% | 0/400.0% | 4/4010.0% | 2/4 | first run |
| 20 | Appsmith | 5 of 1603.1% | 2/405.0% | 0/400.0% | 0/400.0% | 3/407.5% | 2/4 | first run |
| 21 | Hostinger Horizons | 5 of 1603.1% | 1/402.5% | 1/402.5% | 1/402.5% | 2/405.0% | 4/4 | first run |
| 22 | Wix | 5 of 1603.1% | 2/405.0% | 1/402.5% | 1/402.5% | 1/402.5% | 4/4 | first run |
| 23 | Base44 | 3 of 1601.9% | 0/400.0% | 1/402.5% | 2/405.0% | 0/400.0% | 2/4 | first run |
| 24 | Amazon Q Developer | 2 of 1601.2% | 1/402.5% | 1/402.5% | 0/400.0% | 0/400.0% | 2/4 | first run |
| 25 | Augment Code | 2 of 1601.2% | 0/400.0% | 0/400.0% | 2/405.0% | 0/400.0% | 1/4 | first run |
| 26 | Squarespace | 2 of 1601.2% | 1/402.5% | 1/402.5% | 0/400.0% | 0/400.0% | 2/4 | first run |
| 27 | Devin | 1 of 1600.6% | 0/400.0% | 0/400.0% | 0/400.0% | 1/402.5% | 1/4 | first run |
| 28 | Rork | 1 of 1600.6% | 0/400.0% | 0/400.0% | 1/402.5% | 0/400.0% | 1/4 | first run |
| 29 | Cline | 1 of 1600.6% | 1/402.5% | 0/400.0% | 0/400.0% | 0/400.0% | 1/4 | first run |
| 30 | Tabnine | 1 of 1600.6% | 0/400.0% | 1/402.5% | 0/400.0% | 0/400.0% | 1/4 | first run |
What this measures: 4 AI engines · 40 buyer questions each · recommendations counted monthly. Engines = how many of the 4 recommended the tool at least once.
Cursor is recommended about twice as often as Replit: 63 answers to 31 out of 160.
Mentioned at least once but never recommended: OpenAI Codex, Budibase, Create, Firebase Studio. Every other tool on our roster did not come up in a single answer.
What stands out
- 01
When someone asks an AI how to build an app, Cursor comes up in about 1 of every 3 answers. It was recommended in 63 of the 160 answers we collected, more than any other tool, and named in 73.
- 02
Cursor is recommended about twice as often as Replit: 63 answers to 31 out of 160.
- 03
Claude Code is why this page is split by intent. Developers hear about it constantly: recommended in 11 of the 28 developer answers, 3rd in that category. Non-coders almost never do: 5 of 88. Averaged together it lands 11th overall, which reads like weakness and is not. It is a terminal tool for people who already code, and the engines steer beginners elsewhere.
- 04
The category is top heavy. Cursor, Lovable, Bolt took 148 of the 474 recommendations the 4 assistants made this month, 31% of all of them.
- 05
Windsurf shows the gap between visibility and endorsement: named in 23 answers but recommended in only 9. Being known is not the same as being picked.
Where the engines disagree
The overall ranking hides real splits. The same question gets a different answer depending on which assistant your buyer happens to use, and these are the widest splits this month.
- Cursorspread 30.0 pts
Grok recommends Cursor in 23 of its 40 answers. ChatGPT does in only 11.
ChatGPT11/40Claude15/40Perplexity14/40Grok23/40 - Lovablespread 25.0 pts
Grok recommends Lovable in 16 of its 40 answers. ChatGPT does in only 6.
ChatGPT6/40Claude11/40Perplexity14/40Grok16/40 - FlutterFlowspread 25.0 pts
ChatGPT recommends FlutterFlow in 14 of its 40 answers. Perplexity does in only 4.
ChatGPT14/40Claude7/40Perplexity4/40Grok6/40 - v0spread 22.5 pts
Grok recommends v0 in 14 of its 40 answers. ChatGPT does in only 5.
ChatGPT5/40Claude7/40Perplexity6/40Grok14/40 - Boltspread 17.5 pts
Grok recommends Bolt in 13 of its 40 answers. ChatGPT does in only 6.
ChatGPT6/40Claude11/40Perplexity8/40Grok13/40
Developer tools only
Recommendation vs capability
Being good and being recommended are different things. For the developer category we put each tool's share of our 28 developer answers next to its best published score on Terminal-Bench 2.1, a public benchmark that scores the shipping agent itself, tool plus model, on real terminal work. Where the two disagree, that gap is the story.
AI recommendations
11 of 2839.3%
developer answers, August 2026
Terminal-Bench 2.1
83.8%± 1.2
Claude Code with Fable 5, June 7, 2026
Recommended less than its benchmark suggests. The top score on the board gets it third place in the developer recommendations.
AI recommendations
0 of 280.0%
developer answers, August 2026
Terminal-Bench 2.1
83.1%± 1.1
Codex with GPT-5.5, May 1, 2026
The sharpest divergence we found. Statistically tied with the leader on the benchmark, and not one of the developer answers named it.
AI recommendations
23 of 2882.1%
developer answers, August 2026
Terminal-Bench 2.1
79.3%± 1.5
Cursor CLI with Grok 4.5, July 9, 2026
Recommended more than benchmarks alone explain. It trails Claude Code and Codex on the leaderboard and still takes the most developer recommendations by a wide margin.
AI recommendations
0 of 280.0%
developer answers, August 2026
Terminal-Bench 2.1
65.8%± 1.4
Gemini CLI with Gemini 3 Pro, May 1, 2026
The lowest score of the four and the same silence in the answers. Here the benchmark and the recommendations at least point the same way.
Benchmark scores are the best published entry per tool on the Terminal-Bench 2.1 leaderboard, read on 2026-08-21; the date next to each score is that entry's submission date. GitHub Copilot, Windsurf, Aider, Continue, v0, Amazon Q Developer, Augment Code, Cline, Replit, Tabnine also earn developer recommendations but have no published entry on this leaderboard, so they appear without a benchmark number rather than with an invented one.
For toolmakers
Put your ranking on your site
The top 8 tools by share of voice get an embeddable badge. Paste the snippet into your site: the badge links back to this index and we update the underlying data monthly, so it never goes stale. Swap -light for -dark in the URL for the dark variant.
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/cursor-light.svg" alt="Cursor is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/lovable-light.svg" alt="Lovable is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/bolt-light.svg" alt="Bolt is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/bubble-light.svg" alt="Bubble is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/v0-light.svg" alt="v0 is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/replit-light.svg" alt="Replit is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/flutterflow-light.svg" alt="FlutterFlow is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
<a href="https://claw.mobile/ai-recommendations"><img src="https://claw.mobile/badges/glide-light.svg" alt="Glide is recommended by AI assistants (August 2026) per the claw.mobile AI Recommendation Index" width="300" height="60" /></a>
Methodology
The point of this index is that anyone can check it. Same questions, same neutral prompt, every month, and the raw dataset is downloadable below.
Questions
A fixed set of 40 questions (set v1), written the way a real buyer types them: category queries like "best AI app builder", head-to-heads like "Bolt vs Lovable", jobs to be done like "build an internal tool from a spreadsheet", and per-tool checks like "is Lovable worth it". The set is frozen so months are comparable. Since August 2026 each question also carries a buyer-intent tag (non-coder, developer, specific build) that powers the category boards above; the tags are metadata on the same frozen set, so no question text changed and the version did not bump.
Engines and models
Each question was asked once per engine through the official APIs on 2026-08-20: ChatGPT (gpt-5.2), Claude (claude-sonnet-5), Perplexity (sonar), Grok (grok-4-latest). Default parameters, and the only system prompt is "Answer as you normally would", identical for every engine, so none is steered toward or away from any vendor.
Counting
Each answer is classified against a fixed roster of 42 tools. "Mentioned" means the tool is named at all. "Recommended" means the answer presents it as a pick for the asker: it tops a list, is called best for the use case, or is explicitly suggested. A tool named only in passing or as a warning counts as mentioned, not recommended. Grok (xAI) became the fourth engine in August 2026 and its answers are counted above. Grok the tool joined the roster at the same time; a tool is only ranked once every engine's answers are classified against it, so Grok appears in the tool rankings from the September run.
Failures
All 160 planned answers came back. Any call that fails after retries is excluded from the denominators, and a partial run is labeled as such.
Raw dataset: 2026-08.json (CC BY 4.0, cite claw.mobile). Every raw answer is archived, so any number on this page can be audited back to the exact answer text.
If your tool should be in these answers and is not, that is a fixable problem, and it is what we work on with vendors. Work with us. The index itself is not for sale: paying changes nothing in the numbers above.