ChatGPT, Perplexity and Gemini for SEO tasks

This guide is part of our "SEO tools explained" path. It's for site owners, freelancers and small agency teams who already use an AI assistant now and then and want to use it properly for SEO work, without publishing mistakes or leaking client data.

It's about using AI assistants as tools. Getting your own pages cited inside AI answers is a different job, and our GEO guide covers it.

By the end you'll know which assistant to open for which job and which privacy settings to change on day one. You'll also be able to run about ten real SEO tasks, each with a prompt you can copy and a check to run afterwards. The tasks run from keyword brainstorming and clustering to briefs, titles, schema, regex, spreadsheet formulas, crawl and log data, translation and cited research. Our examples are made up: a sweet shop in Kanpur, a dentist in Delhi, a coaching centre in Pune and a D2C skincare brand.

There are seven chapters, about 40 minutes in all. Chapter 1 compares the three assistants and says when you don't need one. Chapter 2 is set-up and privacy. Chapters 3 to 6 are the tasks, with the hallucination risks and what not to do in chapter 6. Chapter 7 has the plans and prices, read on each vendor's own page on 18 October 2026, in rupees where the page showed them. It ends with pros and cons and the alternatives, our free tools first.

Most people who do SEO already have one of these open in a tab. The question isn't whether to use them. It's which one to open for which job, and what to check before anything they wrote reaches a live page.

One line on scope first. This guide is about AI assistants as tools for your own SEO work. If you want your pages to be cited inside ChatGPT, Perplexity or Google's AI answers, that's our GEO guide, and we don't repeat it here.

The three assistants in one paragraph each

ChatGPT (OpenAI) is a general assistant. Its free plan gives unlimited text chats, subject to abuse guardrails, with limited file uploads, data analysis, deep research and memory.^[1] For SEO the useful parts are the chat itself and the ability to read a CSV you upload and work on it. OpenAI said in February 2026 that ChatGPT had 900 million weekly users.^[2]

Perplexity is built as an answer engine. Its pricing page says: "Every answer includes inline citations from trusted sources, so you can verify claims".^[3] That claim is the vendor's. The sources are there to open, but they aren't always right, as chapter 6 shows. The free plan gives "practically unlimited basic searches" and three Pro Searches a day.^[4]

Gemini (Google) runs on your Google account. The free plan includes Deep Research, Canvas and Gems, and 15 GB of storage.^[5] The paid Google AI Pro plan adds Gemini in Gmail, Docs and Sheets, which matters if your keyword lists and reports already live in Sheets.^[6] Sundar Pichai said in July 2026 that the Gemini app had 950 million monthly users.^[7]

Good at, bad at

That table is our opinion from how each product is built and what its pages say, not a benchmark. Answers also change with the model, the plan and the day, so treat the ratings as a starting point.

What every one of them gets wrong

Generative models write the most likely next words. Google's own guidance puts it plainly: they "don't retrieve facts, but predict a likely sequence of words based on their training data", and their output "may contain inaccuracies (also known as hallucinations)".^[8] OpenAI says the same about its own product: "ChatGPT also hallucinates."^[9]

For SEO that means three things you can't take from a chat window. Search volumes and difficulty scores, which assistants make up. Facts, prices, laws and statistics, which need a source you've opened. And quotes, which are often invented. Chapter 6 has the study numbers.

What Google says about AI-assisted content

Google doesn't ban AI help. In February 2023 it wrote: "Appropriate use of AI or automation is not against our guidelines."^[10] What it asks for is people-first content: made to help a reader, with real experience behind it, and clear about who made it and how.^[11]

Guide 2 explains the rest of Google's position, including scaled content abuse and what happened to sites that mass-produced AI pages, in its chapter on E-E-A-T, helpful content and AI-assisted writing. Read it before you publish anything an assistant drafted.

When you don't need one

You don't need an AI assistant to do good SEO. For a 15-page site for a Delhi dental clinic, the jobs that move rankings are the ones only the clinic can do: real fees, real photos, the dentist's name and qualifications, reviews from real patients. An assistant can't supply any of that.

Where it saves real time is the glue work: turning 400 Search Console queries into groups, writing a regex you'd otherwise look up, drafting ten title options, fixing a broken JSON-LD bracket. If you never do that kind of work, the free plan of whichever assistant you already use is plenty.

Next, chapter 2: the settings to change before you paste anything into one of them.

Set-up takes ten minutes. Sign in with the account you'll use for work, change one privacy setting in each assistant, and decide what client data is allowed in a chat at all. Do it before the first client file goes in, not after.

Training and history settings, side by side

The rows come from each vendor's help pages: OpenAI's data controls and temporary chat pages,^[12] Google's Gemini Apps Privacy Hub (updated 13 October 2026),^[13] and Perplexity's privacy and incognito help pages.^[14] Google's plans page says Workspace customers get Gemini through an add-on, not a personal Google AI plan.^[6] The detail for each assistant follows.

ChatGPT

Open Settings, then Data controls, and turn off "Improve the model for everyone". OpenAI says that when it's off, "your new conversations won't be used to train OpenAI models", and they still appear in your history. Signed in, the choice follows your account across devices.^[12] The opt-out only covers new chats, so do it on day one.

Two catches. First, if you rate an answer with thumbs up or down, "the entire conversation associated with that feedback may be used to train our models", even after you've opted out.^[15] Second, turning training off doesn't delete anything. Delete chats separately.

A temporary chat stays out of your history, doesn't create memories and isn't used to improve the models while it stays temporary. OpenAI "may keep a copy of a temporary chat for up to 30 days for safety purposes".^[16] Choose Unpersonalized before your first message if you don't want your memories or custom instructions mixed into a client job.

On ChatGPT Business, Enterprise and Edu, OpenAI says it doesn't use your inputs or outputs to improve its models by default.^[15] An agency with several people on client data should look at those plans rather than personal accounts.

Gemini

Gemini's control is called Keep Activity, managed in Gemini Apps Activity. With it on, Google uses your activity to improve its services, "including training generative AI models", and some chats are read by human reviewers. Activity auto-deletes after 18 months by default; you can choose 3 or 36 months, or no auto-delete.^[13]

With Keep Activity off, future chats aren't used to train Google's models unless you send feedback, but they're still kept for 72 hours. Temporary chats get the same 72 hours and aren't used for training.^[13] Two details matter for SEO work. Chats already seen by reviewers are kept for up to three years, even if you delete your activity. And with Keep Activity off, Connected Apps such as Google Workspace aren't available, so Gemini can't reach your Drive files.^[13]

Google's own advice on the same page: "Please don't enter confidential information that you wouldn't want a reviewer to see".^[13]

Perplexity

Perplexity says it uses search queries and feedback "to improve the search experience". To opt out, go to Account settings, Preferences and turn off AI data retention; Perplexity says it's on by default for Free, Pro and Max users.^[42][14] Perplexity's own pages don't agree on who gets that switch. The privacy help page gives it with no plan limit, while the plan comparison page lists Data Privacy as "Standard" for Free and "Opt-out AI training available" for Pro, Education Pro and Max.^[4] If you don't see the toggle on a free account, assume your searches can be used. It also says its agreements with model providers such as OpenAI and Anthropic stop them training on Perplexity data.^[17]

Incognito mode is Perplexity's private chat; its help page says "All incognito sessions expire within 24 hours and are not recoverable."^[18] Account data is kept while the account is active and removed within 30 days of deleting it.^[19]

What never to paste

A setting is not a contract. Even with training off, chats are stored for a while and some may be read for safety. So keep these out of every chat, on every plan:

• Customer names, phone numbers, emails and addresses, which often sit in form exports and WooCommerce order files
• Passwords, API keys, Search Console or hosting logins
• IP addresses from server logs (remove the column before you upload a log, chapter 5)
• A client's unpublished revenue, prices or launch plans, unless the client has agreed
• Anything under an NDA

For SEO you rarely need any of it. A query export, a crawl export with URLs and status codes, or a list of page titles is enough for almost every task in this guide.

Next, chapter 3: keyword brainstorming and clustering, and where the volumes really come from.

Brainstorming is where assistants earn their place. A person running out of ideas after twenty keywords can get eighty more in a minute. The catch is that some of those eighty nobody searches, and any number attached to them is invented.

What an assistant adds, and what it can't

It's good at the shape of searches: "price", "near me", "online order", "for diabetics", "Diwali gift box", the questions people ask before they buy, and the words a customer uses instead of the trade term. It can also suggest how people in India mix English with transliterated words in searches, which Guide 2 explains in How India searches.

It can't tell you how many people search any of it. Assistants don't have Google's query data. Ask for volumes and you'll get round, confident numbers that don't come from anywhere. Guide 1's rule stands: never take a number from a chatbot; check it in Keyword Planner or Search Console.

Keyword Planner gives "estimates on the number of searches a keyword gets each month". Google says you need to finish the Ads account set-up, including billing information, to use even its basic features, and that keywords with very low search volumes can't be found or forecast.^[21] Our Keyword Planner guide shows how to open it without running an ad and how to read the ranges.

A brainstorming prompt that works

Give the assistant the business, the city, the customer and the season. Ask for phrases, not numbers, and ask it to say which ones it's unsure about.

I run a sweet shop in Kanpur that sells mithai, dry fruit boxes and
sugar-free sweets, with local delivery and courier across India.
List 60 search phrases a customer might type into Google before buying from us.
Group them as: product, occasion (Diwali, Raksha Bandhan, weddings),
local (Kanpur areas), gifting and corporate, and questions.
Use the words customers use, including common Hindi words written in English.
Do not give search volumes or difficulty. Mark any phrase you think is rare with (rare).

The last two lines matter. Telling it not to give volumes stops the made-up numbers, and the "(rare)" mark gives you a first filter, though you still check every phrase.

Clustering: grouping keywords into pages

Clustering means deciding which keywords one page should target and which need their own page. Assistants group by meaning. Google, in effect, groups by results: if two phrases show mostly the same pages, one page can usually rank for both. Those two methods agree most of the time and disagree on the cases that matter, so use the assistant for the first pass and the results page for the close calls.

Upload the cleaned CSV from the task above (keyword and volume only) and ask:

This CSV has keywords and Keyword Planner volume ranges for a Kanpur sweet shop.
Group the keywords into clusters where one page could answer all of them.
For each cluster give: a suggested page (existing type: product, category,
gift guide, blog post or location page), the main keyword, and the others.
List separately any pairs you are unsure should share a page, and say why.
Do not change, add or estimate any volumes.

Then check the "unsure" pairs by hand. Search both phrases and compare the top ten. Lots of the same URLs: one page. Different kinds of pages: two pages. Guide 2's chapter on topic clusters and cannibalisation has the full method, and how to spot two of your pages competing for one search.

That last check catches a common problem: the assistant quietly drops or adds rows. If you sent 312 keywords, the clusters should hold 312.

Next, chapter 4: briefs, outlines, titles, meta descriptions and schema drafts, and the tools to check them in.

This is the work most people use AI for. It's also where small mistakes go live most often: a fee that changed, a doctor's qualification the assistant guessed, a review rating in the schema that nobody gave. The fix is simple. You supply the facts; the assistant supplies the wording.

Briefs and outlines

A brief tells a writer what a page must cover and for whom. Guide 2's chapter on content briefs and search intent explains what goes in one. An assistant can turn your notes and the questions you found in chapter 3 into a tidy brief. It can't see what ranks today unless it searches, and even then it reads only a few pages.

Write a content brief for a page on "root canal cost in Delhi" for a dental
clinic in Lajpat Nagar. Audience: adults with tooth pain comparing clinics.
Use only these facts: [paste the clinic's real fees, what's included,
number of sittings, the dentist's name and qualifications, payment options].
Include: search intent in one line, the questions the page must answer,
an H2/H3 outline, what proof to show (photos, case notes), and what
NOT to claim. Mark any point you added that isn't in my facts with [VERIFY].

The "[VERIFY]" instruction makes the assistant show its own additions. Search the draft for it and check each marked point against the clinic's records, or delete it. For a medical page, the dentist reads the final draft. That's the trust part of E-E-A-T, and no tool replaces it.

If you write many briefs a month, a paid content tool reads the ranking pages for you and lists headings and questions. Our guides to Surfer and Frase show what they add over a chat window, and what they cost.

Titles and meta descriptions

Ask for options, not one answer, and give the rules. Guide 2's chapter on titles, descriptions and snippets has the full rules; here is a prompt that builds them in.

Write 8 title tags and 5 meta descriptions for this page.
Page: root canal treatment, Lajpat Nagar clinic, Delhi. Main phrase: root canal cost in Delhi.
Facts: [fee range, sittings, payment options, dentist's name].
Rules: main phrase near the start; titles under 60 characters;
descriptions under 160 characters; no "best", "top" or "No. 1";
no fees or promises that aren't in the facts. Show the character count after each.

Don't trust the character counts the assistant prints; they're often off. Count them yourself, or check the live page.

The analyzer reads pages from their URL, the way a crawler does, so it checks what's actually published, not a draft in a chat window. Google may still rewrite your title or description in its results; a good tag makes that less likely, but nobody controls it.

Schema (JSON-LD) drafts

Assistants write decent JSON-LD and are very good at fixing broken JSON. They're also happy to invent fields: a star rating, an opening time, a price range or a founding date. Every field must match what the page shows. Google's LocalBusiness documentation says to use the most specific subtype,^[22] and schema.org has a Dentist type for exactly this kind of page.^[23]

Write LocalBusiness JSON-LD using the schema.org type Dentist for this clinic.
Use only these values: [name, full address, phone, opening hours per day,
URL, logo URL, geo coordinates if known].
Do not add aggregateRating, review, priceRange or any field I haven't given.
Output only the JSON-LD script block.

Two more points before you paste FAQ markup an assistant suggests. Google stopped showing FAQ rich results on 7 May 2026, so FAQPage markup won't get you the extra lines in results any more.^[24] And no markup guarantees a rich result.

Alt text works the same way. An assistant can describe a product photo in one line, but it doesn't know that the box in the photo holds 500 grams of kaju katli. Check alt text like any other copy.

Next, chapter 5: the data jobs, from Search Console regex to crawl and log exports.

These are the jobs where an assistant saves the most time and does the least harm, because the output is easy to test. A regex either matches the right rows or it doesn't. A formula either gives the right total or it doesn't. You'll know in a minute.

Regex for Search Console filters

Search Console's Performance report has a Custom (regex) filter for queries and pages. Google's help page says "The RE2 syntax is used." A pattern matches anywhere in the text unless you anchor it with ^ or $. It isn't case-sensitive unless you start it with (?-i). And "Doesn't match regex" excludes instead of including.^[26]

That RE2 detail is where AI-written patterns fail. RE2 doesn't support lookahead, lookbehind or backreferences; the syntax page marks each one "NOT SUPPORTED".^[27] Assistants like lookaheads, so say so in the prompt.

Write a regex for the Google Search Console query filter (RE2 syntax:
no lookahead, lookbehind or backreferences).
Match queries about fees or cost for JEE or NEET coaching, in either order,
for example "neet coaching fees pune" and "fees of jee classes".
Give the pattern on one line, then 5 queries it should match and 5 it shouldn't.

A pattern that works in RE2 looks like this:

(fee|fees|cost|price|charges).*(jee|neet)|(jee|neet).*(fee|fees|cost|price|charges)

Spreadsheet formulas

Ask for the formula, not the answer. If you paste 2,000 rows and ask "which queries are local?", you get a summary you can't audit. If you ask for a formula, you get something you can see working on every row.

In Google Sheets, column A has Search Console queries for a coaching centre in Pune.
Write a formula for column B that labels each query "Local" if it contains
pune, kothrud, baner, wakad, hinjewadi or "near me", "Brand" if it contains
our brand name [name], and "Other" otherwise. Brand should win over Local.

You'll get something close to this, which you can read and correct:

=IF(REGEXMATCH(LOWER(A2),"brandname"),"Brand",IF(REGEXMATCH(LOWER(A2),"pune|kothrud|baner|wakad|hinjewadi|near me"),"Local","Other"))

Fill it down, then read ten labelled rows by eye, including one query with both your brand and an area in it. That one should say Brand.

Google Sheets also uses RE2 for REGEXMATCH, except Unicode character class matching, and the function only works on text, not numbers.^[28] So the same "no lookahead" rule applies. Other formulas worth asking for: XLOOKUP to join a Search Console export with a Keyword Planner export on the keyword column, and a pivot table set-up to total clicks by label.

On a Google AI Pro plan, Gemini works inside Sheets itself.^[6] ChatGPT's free plan has limited data analysis of uploaded files, with more on paid plans.^[1] Either way, check the totals: the sum of clicks in your labelled sheet should equal the sum in the original export.

Crawl exports

A crawl export from Screaming Frog or Sitebulb has URLs, status codes, titles, canonicals and word counts, usually nothing personal, so it's safe to share. Assistants are good at the first sort: group the 404s by folder, list pages with duplicate titles by template, find pages whose canonical points somewhere else.

This CSV is a crawl export of a WordPress site (columns: Address, Status Code,
Title 1, Canonical Link Element 1, Indexability). Group the problems by URL
pattern or template, not page by page. For each group, give the count,
three example URLs, and a one-line fix a developer can act on.
Don't guess at causes you can't see in the data; say "needs checking" instead.

Then spot-check. Take three URLs from each group and test them in our HTTP Status Checker, which follows every redirect hop and shows the response headers for up to 10 URLs at once. Our guides to Screaming Frog and Sitebulb show how to get these exports.

Server log files

Logs are different. Every line holds a visitor's IP address, and a busy site's monthly log can be far too big to upload anyway. Two safer ways to use an assistant:

• Ask for a command or script you run on your own computer. The data never leaves your machine.
• Or remove the IP column first, cut the file to the bot lines you care about, and upload only that.

Write one command for macOS or Linux that reads an Apache access log
(combined log format) called access.log and counts requests per day
from user agents containing "Googlebot". Print date and count, oldest first.
Don't print IP addresses. Explain each part of the command in one line.

Before you trust the counts, check one day by hand: open the log, search that date for "Googlebot" and see whether the number roughly matches.

Either way, a user agent that says Googlebot can be faked. Guide 3's chapter on crawling at scale and log files explains how to verify real Googlebot hits before you draw conclusions.

Next, chapter 6: research with citations, translation, and the things an assistant should never do for you.

The jobs in chapters 3 to 5 are easy to test. The ones in this chapter aren't. A wrong fact in a competitor summary or a clumsy line in a Marathi page looks fine until a customer reads it. So this chapter ends with the rules and a short checking routine.

Competitor research with citations

Perplexity shows numbered sources under each answer, which makes it the natural first stop for "what are competitors saying?" questions. On the free plan you get three Pro Searches a day and one Research query a month; the rest are basic searches.^[4]

I run a D2C skincare brand in India selling sunscreen online.
Which Indian sunscreen brands sell directly on their own websites, and what
do their product pages claim (SPF, PA rating, price for 50 g, free shipping
threshold)? Use only the brands' own websites as sources, give the URL for
each claim, and say "not found" where a page doesn't state something.

Asking for "the brands' own websites" and "not found" pushes the answer away from listicles and guesses. Then open every URL. Check that the page exists, that it says what the answer claims, and when it was updated. Prices on D2C sites change weekly, so write the date next to each one.

What this is good for: a first list of competitors, the claims they lead with, the questions their pages answer, the topics their blogs cover. What it isn't good for: traffic, rankings or keyword data. For those, use Search Console for your own site, and a suite such as Semrush or Ahrefs for others.

How often cited answers are wrong

The biggest test we know of is a study coordinated by the European Broadcasting Union and led by the BBC, published on 22 October 2025. Journalists from 22 public service media organisations in 18 countries, working in 14 languages, rated more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity to news questions.^[29]

Gemini did worst, with significant issues in 76% of its answers.^[29] Two limits: these were news questions, not SEO tasks, and India wasn't one of the 18 countries. But sourcing was the most common failure, and that's exactly the part people trust when an answer shows links.

OpenAI explains why this happens. Its September 2025 research post argues that models hallucinate "because standard training and evaluation procedures reward guessing over acknowledging uncertainty". In one example, a chatbot asked for one of the paper's authors' PhD dissertation title gave three different answers, "none of them correct".^[9]

An Indian example shows what it costs when nobody checks. The Bengaluru bench of the Income Tax Appellate Tribunal passed an order in Buckeye Trust (ITA No. 1051/Bang/2024) dated 30 December 2024 and recalled it on 7 January 2025. In a later Karnataka High Court petition, the trust argued that the order "appeared to have been driven by Artificial Intelligence", according to ITATonline's case digest.^[30] A press report in Law Trend (28 February 2025) said the order rested on judgments that didn't exist, which were "reportedly generated by ChatGPT".^[31] The report doesn't say who it got the ChatGPT link from, so treat that part as a press report, not a finding. If invented case law can reach a tribunal order, an invented statistic can certainly reach a blog post.

Translation and localisation

Assistants give a decent first draft in Hindi, Marathi, Tamil, Bengali and other Indian languages, and they're fast. They also miss what makes a page local: the words customers actually use, prices and units, festivals, and phrasing that sounds natural rather than translated.

Google's rules point the same way. Its spam policy lists, as scaled content abuse, scraping content to generate many pages "including through automated transformations like synonymizing, translating, or other obfuscation techniques" where little value is provided to users.^[32] Its multilingual guidance asks for one language per page, warns against side-by-side translations, and says translating only the boilerplate while the main content stays in one language can create a bad experience.^[33]

A translation prompt that keeps the facts fixed:

Translate this sunscreen product page from English into Marathi for a customer
in Pune. Keep the brand name, product names, SPF and PA values, prices and
pack sizes exactly as they are. Use everyday Marathi, not formal or literary
words, and keep common English words customers use (sunscreen, gel, SPF).
After the translation, list every phrase you found hard to translate and
every place you changed the meaning, with the English original.
[paste the page text]

The list at the end is what your reviewer starts with. Then compare every number in the Marathi version with the English page.

What not to do

Fake reviews deserve their own line. Google's Maps content policy removes "Content that is not based on a real experience" and reviews "paid for, directly or in kind".^[34] In India, the Department of Consumer Affairs launched IS 19000:2022, a Bureau of Indian Standards framework for collecting, moderating and publishing online consumer reviews, in November 2022.^[35] AI makes fake reviews cheap to write. It doesn't make them any less fake.

And scale: Google's spam policies name "Using generative AI tools or other similar tools to generate many pages without adding value for users" as an example of scaled content abuse.^[32] Guide 2's chapter on AI-assisted writing shows what happened to sites that tried it.

A checking routine for anything an assistant wrote

1. Facts and numbers: every one has a source you opened, with a date. Delete what you can't source.
2. Quotes: every one is word for word from a page you read.
3. Tags and schema: titles and descriptions checked on the live page, JSON-LD validated and matched to what the page shows.
4. Language: a person who knows the subject, and for translations a fluent speaker, read the final text.
5. Scale: each page exists because a customer needs it, not because a keyword list had a row for it.

Next, chapter 7: what each assistant costs in rupees, the pros and cons, and the alternatives.

The plans, checked 18 October 2026

We read each pricing page from India on 18 October 2026. ChatGPT and Google showed rupee prices. Perplexity showed US dollars only, so its rupee figures are ours, at ₹96 to US$1, the X-Rates average for 1 to 16 October 2026 that we use across these guides.^[36] None of the three pages mentioned GST or other taxes.

From ChatGPT's pricing page.^[1] Its FAQ says Go and Plus are monthly plans, Business starts at 2 users, and Go, Plus, Pro and Business can be bought "with any major credit card".

From Google's Gemini subscriptions page and Google One's AI plans page.^[5] Only personal Google Accounts can sign up for Google AI plans, and the Gemini features in them are for ages 18 and over; Workspace customers buy a Gemini add-on instead.^[6]

From Perplexity's Pro page and its plan comparison help page.^[3] The Pro page shows only the price per month when billed yearly.^[4] Billed monthly, Pro is $20 a month^[40] and Max $200 a month, or $2,000 a year.^[41]

If you claimed Perplexity Pro free through Airtel's 2025 offer, check your renewal. New redemptions ended on 16 January 2026, the free year runs from each user's activation date, and the subscriptions were set to auto-renew, TechCrunch reported in August 2026.^[37]

Pros and cons

Alternatives, and which one when

Our own free tools are where we'd send the output of an assistant to be checked. Two of them use AI themselves. The Free Website Audit's AI visibility check asks AI who it recommends near you, alongside AI crawler access and llms.txt, as part of 50+ checks. With no account you get the score, every check and live progress, one audit a day; a free sign-in adds every fix with its points, the PDF report and a competitor compare, three audits a day.

The Background Remover cuts out product photos with an AI model that runs in your browser, so the photos are never uploaded, up to 10 images at a time.

Then the checkers this guide used. The Meta Tag Analyzer checks AI-drafted titles and descriptions, and the Schema Markup Validator checks AI-written JSON-LD. The Word Counter & Keyword Density Checker shows a draft's keyword use. The AI Crawler Checker tells you whether the crawlers behind ChatGPT, Perplexity and Gemini may read your site; Guide 5's chapter on technical access explains what to allow.

Surfer's lowest plan is $59 a month billed monthly or $49 billed yearly (about ₹5,664 or ₹4,704), from its pricing page on 18 October 2026.^[38] Frase Starter is $49 a month or $39 billed yearly (about ₹4,704 or ₹3,744), from its pricing page the same day.^[39] Claude and Microsoft Copilot do similar jobs; we haven't covered their plans here.

If you'd rather hand the work to a team, see our content marketing and SEO services. Otherwise, start with chapter 2's settings today, then run the keyword task in chapter 3 on your own site.

Test yourself

Ten questions on the whole guide. Pick an answer, check it, and see your score at the end.

Frequently asked questions

Sources