Original research
How AI sees small businesses in Barcelona
We analysed 240 websites of small businesses in Barcelona, across eight sectors, with the engine behind our free “How does AI see you?” tool. We measured how easy each website makes it for ChatGPT, Gemini, Claude or Perplexity to find and understand the business. Without asking any AI: everything comes from what can be checked on the website itself.
Gerard Bori · SimplicallyKey figures
- 58/100is the average score: the estimated probability that AI finds and understands these websites. Only 7% reach 80 and more than half (54%) score below 60.
- 48%have no business details in structured data (schema.org): AI has to work out from the page what the business is, where it is and how to contact it.
- 42%don't say which city or neighbourhood they are in, either in the title, the headline or the description.
- 9%close the door to every AI crawler in their robots.txt. Nearly all of them (17 of the 19 we re-read) only let Google and Bing in.
The bottom line: the door is usually open, but inside AI finds little reliable information. The websites score 93/100 for visibility but only 36/100 for accuracy.
Which sectors make it easiest for AI
Health (64), shops (63) and accommodation (62) lead the way; food & drink (53) and estate agents (53) come last. Average score of each sector out of 100, with 30 websites per sector: differences of less than 8 points between two sectors may be down to chance.
- Clinics and health64/100
- Local shops63/100
- Hotels and accommodation62/100
- Beauty, hairdressers and gyms57/100
- Trades and workshops55/100
- Professional firms54/100
- Restaurants, bars and cafés53/100
- Estate agents53/100
How the scores are spread
- High (80-100)7%
- Medium (60-79)39%
- Low (40-59)40%
- Very low (0-39)14%
Where they lose points: the four scores
The estimate is the average of four scores from 0 to 100. Each one groups several checks.
- Visibility93Can crawlers get in and read it? robots.txt, noindex, sitemap, text readable without JavaScript and a title that says what you do and where.
- Accuracy36Are the details reliable? Business details in schema.org, matching contact details, recent dates and llms.txt.
- Structure45Is it easy to understand? FAQs, a clear headline, one page per service, internal links and a description.
- Authority56Are there outside signals? Domain age, Wikidata, Wikipedia, linked profiles, https and up-to-date software.
| Sector | Score | Visib. | Accur. | Struct. | Auth. | No details | With llms.txt |
|---|---|---|---|---|---|---|---|
| Health | 64 | 92 | 41 | 60 | 61 | 40% | 31% |
| Shops | 63 | 93 | 44 | 53 | 62 | 40% | 43% |
| Accommodation | 62 | 96 | 45 | 45 | 61 | 43% | 39% |
| Beauty & gyms | 57 | 91 | 37 | 44 | 53 | 40% | 25% |
| Trades | 55 | 92 | 31 | 45 | 52 | 50% | 21% |
| Professional firms | 54 | 91 | 32 | 41 | 50 | 50% | 7% |
| Food & drink | 53 | 91 | 33 | 33 | 54 | 50% | 23% |
| Estate agents | 53 | 94 | 28 | 38 | 53 | 70% | 3% |
| Total | 58 | 93 | 36 | 45 | 56 | 48% | 24% |
What we found
No business details for AI
48% have no business structured data (schema.org Organization or LocalBusiness) on the home page. Only 8% have complete details, with address, phone number and opening hours.
What to do: Add a LocalBusiness block with name, address, phone number, opening hours and profiles. It's code that isn't visible on the page and that search engines and AI read directly.
One in 11 closes the door to AI
9% of the websites we could read close their robots.txt to every crawler except the ones they name. Nearly all of them (17 of the 19 we re-read) use the same template, which only lets Google and Bing in: that shuts out the crawlers of ChatGPT, Claude, Perplexity and Gemini. Counting those that block only some crawlers too, 11% block at least one. The rest leave the door open.
Websites that block each crawler
- OAI-SearchBot (ChatGPT)9%
- Claude-SearchBot (Claude)9%
- PerplexityBot (Perplexity)9%
- Google-Extended (Gemini)11%
- GPTBot (OpenAI)11%
- ClaudeBot (Anthropic)11%
What to do: Check your robots.txt and whether what is there was a decision or a default setting. Blocking the crawlers that train models is a legitimate choice; blocking the search crawlers keeps you out of the assistants’ answers.
They don't say where they are
42% don't mention any city or neighbourhood in the title, the headline or the description, and 45% don't say both what they do and where. For an assistant answering “where's a good physio in Gràcia?”, that is exactly what it needs to read. On top of that, 11% show fewer than 150 words without running JavaScript, which is how AI crawlers read the page.
What to do: A title such as “Sports physiotherapy in Gràcia, Barcelona” and the main text written into the page itself, not loaded later with JavaScript.
What helps AI, half done
One in 4 (24%) already has an llms.txt file, the summary designed for AI assistants (some SEO plugins now create it automatically). But only 16% answer frequently asked questions on the home page, 27% have no description for search engines and 15% have no sitemap (sitemap.xml).
What to do: Publish a sitemap.xml and a simple llms.txt (what you do, where, services and contact) and add the questions customers always ask, with their answers, to the home page.
Signs of a neglected website
29% of the websites showing any date show none from the past year (neither in the sitemap nor on the page), and 23% of those with a year in the footer show one before 2025. 20% of those that repeat a phone number or postcode don't have them matching across the business details, the page and the legal notice. 16% reveal software that no longer gets security fixes, another 13% load out-of-date WordPress plugins and 6% don't use a secure connection.
What to do: Show that the website is alive (dates, footer year, news), use the same contact details everywhere and keep the content management system and its plugins up to date.
Almost invisible outside their own website
Only 5% are on Wikidata with their official website and 11% have a Wikipedia article (in Spanish, Catalan or English) linking to them. 26% don't link to any social profile. Half of the domains that publish their registration date are more than 14 years old.
What to do: Get others talking about you: profiles linked from your website, sector directories, associations, local press and, if you meet its criteria, a Wikidata entry.
Five things any small business can do this week
1. See how AI sees you
Run your website through the free analyser: in a minute you see what is missing and why it matters.
2. Add your business details
A LocalBusiness block with name, address, phone number, opening hours and profiles.
3. Check your robots.txt
Let the assistants' search crawlers in, if you want to appear in their answers.
4. Say what you do and where
In the title and in the home page text, written into the page.
5. Same details everywhere
The same name, address and phone number on your website, on Google and in directories.
Methodology
Open data and how to cite it
You can use these figures and the data file by sector (CC BY 4.0 licence) by citing the source with a link to this page.
Download the data (CSV)How to cite it
Simplically (2026). How AI sees small businesses in Barcelona. https://simplically.com/en/ai-small-business-barcelona-study/
Business data: © OpenStreetMap contributors, under the ODbL licence.
- Which websites
- Businesses in eight sectors within the municipality of Barcelona with their website recorded in OpenStreetMap (data from 10 November 2026): 3,598 map points. We removed chains and franchises (1,063 map points: with a recorded brand, with the same domain at more than one point in the province of Barcelona, or with an address that is a branch page within a chain's website), social profiles and platforms (TheFork, Booking…) and public-sector websites. That left 2,452 websites, from which we drew 30 per sector at random (fixed seed, 20261110: the draw can be repeated). Even so, a well-known brand with a single location on the map may remain.
- How many
- We visited 332 websites: 240 were analysed, 23 closed the door to all crawlers in their robots.txt (we didn't open them) and 69 could not be analysed (site down, outdated address, server error or anti-bot challenge). Each website that failed was replaced by the next one drawn in its sector. Those that closed the door to all crawlers have no score, but they do count in the crawler figures (19 of them were re-read by hand on 10 November 2026).
- How
- With the same engine as the “How does AI see you?” analyser, run locally, on 10 November 2026. One visit per website and one website at a time, with pauses: the page recorded on the map (almost always the home page), robots.txt, llms.txt, the sitemap, the legal notice and little else, never requesting anything their robots.txt blocks. Outside the website we checked Wikidata, Wikipedia and the domain registry (RDAP). The “is on OpenStreetMap” check doesn't count, because every website comes from there.
- What it does not measure
- It doesn't ask any AI: it's an estimate based on verifiable data, not what each assistant answers today. It doesn't look at reviews, Google profiles, press or directories, and it only analyses one page of each website.
- Limitations
- These are businesses someone has already recorded on the map with their website, so they are probably a little more up to date than average: the real picture is probably worse. The total gives the same weight to all eight sectors (it isn't weighted by how many businesses there are in each). Margin of error (95%): around ±7 points for overall percentages and up to ±18 for each sector. .es domains don't publish their registration date (no data).
- Privacy
- Business websites only. We don't publish or single out any business, and website-level results are not shared or used to contact anyone.
What about your website? Analyse it for free
The same analysis as in the study, for your website, in a minute: score, every factor measured and how to improve it.