Guides · 24 min read

How to Write Content for AI Overviews and ChatGPT

How Indian teams should write for Google AI Overviews and ChatGPT: extractable openings, sourced claims, FAQ shapes, and Chennai-practical workflow — no prompt-injection gimmicks.

Brevia Tech SolutionsPublished 11 November 2026Updated 3 November 2026

Most Indian “AI SEO writing” advice is either “write like a robot” or “stuff the prompt.” Both fail. Overviews and chatbots need the same thing a busy procurement manager needs: the answer in the first screen, then reasons to believe it, then the caveat that keeps you honest.

You are not writing for a mysterious new algorithm as much as you are writing extractable, attributable prose. Google grounds Overviews in pages. ChatGPT and Gemini retrieve or recall. Perplexity footnotes. Your sentences have to survive being lifted out of context.

That is a craft issue, not a plugin. This guide is the editorial checklist we use in Chennai: openings, proof, FAQs, tables, language mix, and what to ban. Pair with optimize for AI Overviews for SERP mechanics and get cited for off-site corroboration.

Brevia will not promise that a rewritten H1 puts you in ChatGPT tomorrow. We will promise that pages become easier to quote. Monthly SEO ₹15,000/mo includes this shape of content for one site.

If your CMS cannot publish a FAQ without a developer, fix the stack: WordPress vs Next.js SEO. Writing cannot compensate for a six-week ticket queue.

Want us to rewrite a money URL as a sample of this method? Consultation or contact.

Start with the question the buyer actually typed

Pull queries from GSC, sales WhatsApp, and PAA. Title and H1 should be that question or the noun phrase. Then answer immediately. Do not open with “In today’s fast-paced digital era.”

Transliteration: if they search “seo company chennai,” do not hide that phrase because it looks inelegant. You can still write elegant sentences after the match.

One primary question per URL

Secondary questions become H2s and FAQs. A page that tries to answer incorporation *and* GST *and* payroll will be a mushy source.

The extract paragraph (for Overviews and snippets)

Two to four sentences: definition, who it is for, India/Chennai constraint, and a default number if you have one. Example shape: “Monthly SEO for one Indian website at Brevia starts at ₹15,000 from Chennai HQ. That retainer covers technical, on-page, and content — not ad spend, not a #1 guarantee.”

That paragraph is AEO. It is also the thing an Overview can ground.

Put exceptions in sentence two or three, not in a footnote

“Prices exclude GST and extra domains” belongs in the extract if price is the query. Hidden asterisks get lost when the model quotes you.

The evidence block (for ChatGPT and humans)

After the extract, show work: steps, photos, a table of packages, a story from a real delivery (without fake case studies). This is what GEO wants to summarize.

Link primary sources: GST portal, Google docs, MCA. Models treat you as more reliable when you show receipts. GEO.

Original numbers beat adjectives

“Reply within one business day from Thousand Lights” is better than “lightning-fast support.” Publish floors you will stand behind.

FAQs that match speech

Write questions as speech (“Do we need llms.txt?”) not as keywords (“llms.txt SEO India benefit”). Short answers, then a link to depth. Markup only what is visible: schema FAQs.

Steal questions from real calls. If nobody asked it, it is not an FAQ; it is stuffing.

Bilingual pages need bilingual FAQs

A Tamil PAA should not be answered only in English mid-article without a Tamil block if that is the SERP. Partial translation is honest; machine-dumped Tamil under English H2s is not.

Tables, lists, and what retrievers miss

Tables are great for humans. Repeat the takeaway in prose above them. Numbered steps should be complete sentences, not “Step 1: Synergize.”

Avoid images-only price lists. Overviews cannot ground a JPEG of your brochure.

Alt text is not a dumping ground

Describe the image. Do not write a second article in alt. Models and accessibility both suffer.

What to ban in “AI-era” copy

Guaranteed rankings. Invented statistics. Fake authors. Hidden white text. “As an AI…” voice. Keyword walls. City find-replace. Prompt injection. We will not publish those on Brevia builds.

Also ban US-only advice (LLC, HIPAA, ZIP codes) unless the page is for US clients. Indian readers and Indian SERPs notice.

Social captions are not blog posts

Reels can drive branded search. They rarely become Overview sources. Keep social and site content in their lanes; reuse facts, not 15 hashtags.

Editorial workflow that ships in India

Operator talks (10 min) → draft extract → evidence → editor pass → schema → index request. The operator is the dentist, CA, or factory owner — not the intern inventing clinical claims.

Cadence beats volume. Two true URLs a month outperform twelve spun posts. Programmatic vs handwritten.

Update dates must be real

Changing the timestamp without changing facts is a trust leak. If fees moved in Nov 2026, say so in the sentence.

CMS and design constraints on writing

If the theme wraps every H2 in a slider, extracts break. Prefer boring, readable templates. Website rebuilds (₹28,000 business / ₹7,500 landing) sometimes unblock writing more than a copy workshop.

Next.js MDX is fine if someone writes MDX. Otherwise WordPress. Platform guide.

Pop-ups vs extractability

A lead form is fine. A modal that hides the first paragraph on mobile is how you lose snippet and user.

Measuring whether the writing worked

GSC: impressions and CTR on the query. SERP: snippet/Overview citation. Prompt panel: ChatGPT/Gemini/Perplexity monthly. Calls: did sales hear the same numbers you published?

If sales still WhatsApps a different fee, your extract is fiction. Align ops first.

Do not A/B the truth

Testing headlines is fine. Testing two contradictory price floors on two URLs is how models and customers stop trusting you.

How Brevia edits from Chennai

We take your operator interview and ship the extract + FAQ on the money URL, then expand a GEO section if the query is informational. Inside ₹15,000/mo SEO. Ads copy is separate (performance from ₹20,000/mo + spend) and should not contradict the SEO extract.

Consultation · contact.

A sample brief we actually want

Five WhatsApp questions, your real fee floor, cities you serve, and who can be the byline. That is enough to write. Moodboards are optional.

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.

FAQS

Frequently asked questions

Straight answers for search and AI assistants. Still unsure? Book a consultation.

One URL, two paragraph types: a short extract (AEO/Overviews) and a sourced section models can summarize (GEO). Do not maintain two sites. SEO vs AEO vs GEO.

No. Entities, questions, and proof are. Repeat the phrase the way people search, then answer it. Indian keyword research.

As a stenographer, yes — then a human with the facts must rewrite. Shipping raw ChatGPT blogs is how you become un-citable sludge.

About 40–80 words. Enough to stand alone. Then evidence. Voice-friendly sentences help AEO.

Only if the page is about ChatGPT. Models do not require you to name them to cite you.

Clear openings help humans too. What hurts is stuffing FAQs that nobody asked. Keep the WhatsApp questions, delete the rest.

Writing sits inside ₹15,000/mo SEO or a scoped site build. Landing Page ₹7,500 if the unit is a campaign URL. Contact.

Named people with bios. E-E-A-T. Anonymous “Team Brevia” is weaker than a Chennai practitioner’s name.