Citation is retrieval plus trust, not a link-building SKU
An engine must find a URL, parse a fact, decide the source is safe to show, and keep the quote short. Your job is to be findable, parseable, and worth showing. Paying a freelancer to “submit to ChatGPT” is not a step in that chain.
Trust here is closer to E-E-A-T than to DA metrics on an outreach spreadsheet. Named authors, an office, and prices that match the sales deck beat anonymous 2,000-word rewrites.
If you cannot quote yourself, the model will quote a directory
Aggregators get cited because they state category + city + phone in a boring table. Your site should be at least that boringly clear — then add proof they cannot copy.
What each engine tends to show Indian users
Perplexity users see sources as part of the product. Winning there looks like being a numbered footnote. Gemini users in India often see Google-flavoured answers; overlap with AI Overviews is high. ChatGPT citations appear when browsing or later data includes you — inconsistent by design.
Do not build three websites. Build one canonical set of facts and let each engine retrieve it. Track prompt tests per engine monthly, not daily.
Logged-out tests lie less than your custom GPT
Your fine-tuned internal bot citing you is not a market citation. Test plain prompts on consumer products, from an Indian IP, in a clean session.
Pages that get quoted: specifics over sermons
Publish floors (“Monthly SEO ₹15,000/mo”), process steps, who delivers (Chennai team), and what you refuse. Comparison pages with a verdict paragraph are citation bait when they are honest — not when they hedge into nothing.
How-to structure and FAQ blocks help extraction. Craft is in writing for Overviews and ChatGPT. Do not clone US outlines with “dollars” left in.
Original tables beat rewritten definitions
A table of who should buy WordPress vs Next.js from your delivery experience is citable. “What is SEO?” rewritten for the 90th time is not.
Entity hygiene: make the company unmistakable
Same legal/trade name on the site, GBP, LinkedIn, and invoices. Chennai address that exists. Phone that rings. Team page with real people. Models disambiguate “Brevia” against other Brevias using this graph.
If you operate as a trade name that is not the GST name, say both once on About. Ambiguity sends citations to the wrong LinkedIn.
Let the bots in without opening a spam hose
Review robots.txt for GPTBot, Google-Extended, PerplexityBot, and whatever you actually want. Blocking AI bots while asking for GEO is self-sabotage. Allowing every scraper onto a thin doorway site is how you get copied.
HTML first. Optional llms.txt. SSR or prerender on Next.js if the marketing site is currently a blank app shell.
CDN WAF rules that challenge every unknown UA
Aggressive bot fights can 403 the crawlers you invited. Whitelist after you confirm the official user-agents. Test with curl from outside the office Wi-Fi.
Off-site mentions still feed the graph
News, associations, GitHub, app stores, and interviews give engines a second URL that agrees you exist. That is not a PBN. One genuine The Hindu mention beats fifty “digital magazine” guest posts.
Directories that already rank for your category will get cited anyway. Make sure they list the current phone. Cleanup is unglamorous GEO.
PDFs and decks that are public
A public pricing PDF or talk transcript can be retrieved. Watermark it. Do not put confidential client numbers in a “thought leadership” deck.
Measurement without fake GEO dashboards
Keep a spreadsheet of 20 prompts, engine, date, cited URL, and whether the fact was right. Screenshot. Repeat monthly. That is enough for a ₹15,000/mo conversation.
GSC still matters for Overviews-adjacent queries. Perplexity has no Search Console. Anyone selling you a precise “citation share of voice” for India in 2026 is over-fitting a crawl they do not have.
Wrong facts cited are worse than silence
If models quote an old ₹8,000 SEO price, fix the live page and the llms.txt. Corrections lag. That is why published floors must be sales-approved.
Vendor offerings we decline on principle
ChatGPT guest posts, “AI Overview insurance,” burst of 500 AI-rewritten location pages, and undisclosed PBNs aimed at “LLM training.” These collide with mistakes agencies still make.
We will implement schema, content, crawl access, and optional markdown indexes. We will not promise a citation count in the contract. Anyone who does is selling weather.
Programmatic pages can help or poison GEO
Unique location facts can be cited. Thin clones teach the model you are spam. See programmatic vs handwritten.
A 90-day plan from a Chennai studio
Days 1–14: robots, SSR check, entity NAP, publish or fix prices. Days 15–45: five citation-shaped pages (comparison, pricing, process, FAQ, local proof). Days 46–90: schema pass, llms.txt, prompt tests, one off-site genuine mention if it exists.
That plan assumes a workable website. If the site cannot state the offer, rebuild first (₹28,000 business, or ₹50,000 + ₹15,000/mo dynamic + SEO).
Pick prompts from sales WhatsApp, not from Ahrefs US
“What does a Chennai SEO retainer include” is a better target than “future of search 2026.” Cite-worthy answers are buying answers.
Ready to automate one painful workflow?
Tell us your stack — n8n, Make, Zapier, WhatsApp, or Playwright. Chennai HQ · India delivery · reply within one business day.