I Asked ChatGPT, Perplexity and Gemini for 'Best Web Agency in Delhi NCR' Every Day for 14 Days. Here's Who Gets Named and Why.

Somebody asked an AI for an agency recommendation this week and your name did not come up. You will never know it happened. There is no search console for ChatGPT, no ranking report, no keyword tracker that catches it. The lead just quietly went to someone else.

That bothered me enough to test it properly.

See, everyone in my space is shouting about "AI SEO" right now, and almost nobody has actually sat down and measured anything. So in May I ran the most boring experiment I could design: ask three AI assistants the same questions every single day for 14 days, write down who gets named, and look for the pattern. No theories first. Log first, theory later.

The setup, because the design matters more than the result

If you only remember one thing from this post, remember the method. Anyone can run this for their own city and their own industry, and you should, because my log tells you about Delhi NCR web agencies and nothing else.

Three prompts, word for word, every day:

1. "Best web agency in Delhi NCR"
2. "I run a small business in Noida and need a website. Which agency should I hire?"
3. "Recommend a web design and development agency in Delhi NCR for a startup, with reasons"

Rules I forced on myself:

• Fresh chat every time. No memory, no logged-in personalization where I could turn it off. ChatGPT got a temporary chat, Perplexity and Gemini got new sessions.
• Same time window each day, between 9 and 10 in the morning.
• Every response went into a Google Sheet: date, model, prompt, every business named, and the order they appeared in.
• 3 prompts, 3 assistants, 14 days. 126 responses total.

That's it. A sheet and stubbornness. The whole thing took maybe 12 minutes a day.

What the log actually shows

First surprise: the assistants name real businesses far more often than I expected. I assumed I'd get hedging ("it depends on your needs...") most of the time. Wrong. ChatGPT gave at least one concrete agency name in 39 of its 42 responses. Perplexity named businesses every single time, because naming sources is its whole personality. Gemini hedged the most and leaned hard on Google Maps style listings for prompt two.

Second surprise: the answers are unstable day to day but stable in aggregate. The exact list changed almost every morning. The pool it drew from did not. Across all 126 responses, the same 6 agencies accounted for roughly 70 percent of every mention in my sheet. Different order, different subsets, same small club.

And yes, since you're wondering: Webight got named twice in 14 days, both times by Perplexity, both in the second week. Two mentions out of 126 responses. I'm writing this post partly because I lost, and losing with a spreadsheet open teaches you more than winning blind.

So I stopped looking at who got named and started looking at what the named agencies had in common. I opened every one of those repeat names and audited them the way I'd audit a client site. Four things showed up again and again.

Pattern 1: their entity data agrees with itself everywhere

Every agency in that club of 6 has the same name, same city, same one-line description on their website, their Google Business Profile, their LinkedIn page, Clutch, and GoodFirms. Boring, identical, repeated.

This matters because an AI assistant deciding whether to name you is doing entity resolution. It has to be confident that "XYZ Studio" on Clutch, "XYZ Studio Pvt Ltd" on LinkedIn, and "xyzstudio.in" are one business before it will stake an answer on you. Every inconsistency lowers that confidence, and a low-confidence entity gets dropped from the answer, silently.

I learned this on my own site the embarrassing way. In our June audit of webight.com, the canonicals pointed to webight.com while the server 307-redirected everything to www.webight.com. Two addresses claiming to be the real one. A human never notices. A machine building an entity graph absolutely does. We fixed it the same week.

Pattern 2: they exist in places that are not their own website

Perplexity cites its sources, so this one was easy to verify. The citations behind its recommendations were almost never the agencies' own sites. They were Clutch profiles, GoodFirms listings, "top 10 agencies in Delhi" roundup posts, and Google reviews.

Your website is your claim. Directories and reviews are the corroboration. Assistants trained to avoid making things up prefer corroborated claims, for the same reason you'd trust a stranger more if two of your friends vouched for him.

The agencies getting named had 30, 50, 80 reviews. Not thousands. Just enough independent signal that recommending them felt safe.

Pattern 3: their pages answer the question in plain sentences

This is the one most people miss. The named agencies all had at least one page that literally answers the buyer's question in extractable prose. Sentences like: "We are a 12-person web development agency in Gurugram. We build websites for D2C brands. Projects start at X and take 4 to 6 weeks."

Subject, verb, fact. A language model can lift that and hand it to a user with confidence.

Now compare that with the average agency homepage. "We craft digital experiences that inspire." Inspire whom? To do what? For how much? There is nothing in that sentence a model can extract and repeat. You wrote a page for an awards jury and then wondered why a machine reading for facts skipped you.

Pattern 4: their HTML tells machines exactly what the business is

Most of the repeat names had Organization or LocalBusiness schema in their HTML. It takes ten minutes. Here is the exact block, ready to paste into a Next.js layout:

// app/layout.tsx (inside <body>, renders JSON-LD for the whole site)
const orgSchema = {
  "@context": "https://schema.org",
  "@type": "ProfessionalService",
  name: "Webight",
  url: "https://www.webight.com",
  description:
    "Two-person design and development studio in Noida, Delhi NCR. Websites, landing pages, and AI automation for clients in India, the Gulf, the US, Southeast Asia and the Netherlands.",
  address: {
    "@type": "PostalAddress",
    addressLocality: "Noida",
    addressRegion: "Delhi NCR",
    addressCountry: "IN",
  },
  founder: { "@type": "Person", name: "Garvit Sharma" },
  priceRange: "INR 7,999 - INR 24,999+",
  sameAs: [
    "https://www.linkedin.com/company/webight",
    "https://medium.com/@sgarvitg",
  ],
};

export function OrgSchema() {
  return (
    <script
      type="application/ld+json"
      dangerouslySetInnerHTML={{ __html: JSON.stringify(orgSchema) }}
    />
  );
}

Swap in your own details, keep the description factual, and make sure every URL in sameAs shows the same name and city as the page it sits on. Pattern 1 and pattern 4 are the same job done in two places.

The checklist to become nameable

Everything above collapses into work you can do this week:

1. Write one canonical sentence about your business: who, where, what, for whom, from what price. Use it verbatim on your homepage, Google Business Profile, LinkedIn, and every directory.
2. Fix your canonical and redirect story so your domain has exactly one address. One. Check it today with curl -I yourdomain.com.
3. Claim Clutch, GoodFirms, JustDial, and Google Business Profile. Then ask your last 10 clients for a review. Most owners have never asked even once.
4. Add a page that answers buyer questions in plain prose: what you cost, how long projects take, who you work with. Facts a model can quote.
5. Add the schema block above.
6. Run the 14-day log yourself for your own niche and city, before and after. The sheet is the only honest scoreboard you'll get for this channel.

None of this is clever. That's the uncomfortable part. The agencies getting named by AI are mostly the ones that did unglamorous data hygiene for years while everyone else A/B tested hero animations.

I'm rerunning the experiment in August after Webight finishes its own checklist. If the mention count doesn't move, I'll write that post too, because a logged failure is still data and a vibe is still nothing.

So before you spend one more rupee on ads, open ChatGPT, ask it the question your customer would ask, and see if you exist. If you don't, you now have the checklist. The only question left is whether you'll do the boring work or keep being invisible in the one channel that's growing.

If you run this for your own industry, send me the sheet. I genuinely want to see if the pattern holds outside web agencies.