Optimization Theory·Field notes

Your SEO
Swiss Army knife.

Connect Claude Code to DataForSEO, add your site data, and turn a question into a keyword plan, a technical fix or a report you can use.

Go to the free SEO Prompt Builder tool

Swiss Army knife with SEO tools, labeled Claude + DataForSEO
The assistant
Claude Code
Runs the work
Claude Pro, $20/mo
+
The search data
DataForSEO
Supplies the evidence
$50 prepaid credit
+
The site context
Your crawl + analytics
Shows what matters
Screaming Frog free to 500 URLs
=
The result
A useful deliverable
Ready for your review
about $45/mo

Claude Pro, about $25 of DataForSEO calls and the free Screaming Frog. With Sitebulb Lite it's about $63. Semrush Guru is $249.95 a month.

Most SEO work involves moving between tools. You find a promising query in Search Console, check the results, open the page, compare it with a crawl and put the recommendation in a spreadsheet. The decision is often simple. Getting everything in front of you takes the time.

Claude Code can take on much of that preparation. Give it DataForSEO and your site data, and you can ask it to investigate a question, collect the evidence and produce something useful: a keyword map, a prioritized fix list, a content brief or a redirect file ready for review.

That's the Swiss Army knife I want more SEOs to know about. I've been doing technical SEO for over 15 years, and what interests me is how much of the work between noticing a problem and preparing the fix now happens in one conversation. Here's how the pieces fit, where I use them, and how to get a useful first result.

Claude and DataForSEO connected by an API pipe, sending search data back and forth and producing spreadsheets, checklists and reports

Claude asks, DataForSEO answers, and the deliverables land on your desk.
Infographic · the whole guide on one pageSEO Expert's Secret Weapon: Claude Code + DataForSEOThe setup, the six APIs, both real runs and your first job, in one image you can save or share.View infographicHide infographic
SEO Expert's Secret Weapon: Claude Code + DataForSEO infographic showing the setup, six DataForSEO APIs, four setup steps, two real runs and a first job prompt

Open full size1024 × 6175 PNG
Share this infographicembed code
<a href="https://optimizationtheory.com/claude-code-dataforseo/"><img src="https://infographi.ac/wp-content/uploads/2026/10/SEO-Swiss-Army-Tool-v3.png" alt="SEO Expert's Secret Weapon: Claude Code + DataForSEO infographic" width="1024" height="6175" style="max-width:100%;height:auto"></a><br>Infographic by <a href="https://optimizationtheory.com/">Optimization Theory</a>
Pasteinto any page or post. Please keep the credit link.

The ingredients

What Claude Code and DataForSEO each bring to SEO

Claude does the work. DataForSEO supplies the evidence. Your crawl and analytics supply the context.

Claude Code is the working environment. It can read files, run scripts, use connected tools and create the output you ask for. DataForSEO supplies external search data: keyword metrics, live search results, ranking estimates, competitor data, backlinks and more through its APIs. Your crawler and analytics fill in the parts that belong only to your website.

This distinction matters. An AI with web access can research current information, but web search doesn't give it a dependable keyword database or your site's performance history. Ask for search volume without naming a source and you may get an unsupported number. Connect a data provider and you can require the actual result, market, collection date and source behind every recommendation.

Three stacked layers: Claude does the work, DataForSEO supplies the evidence, your site crawl and analytics supply the context, feeding a finished deliverable

SERP API

Live Google results for any keyword, country or city: rankings, featured snippets, People Also Ask, AI Overviews.

$0.0006 a SERP queued, $0.002 live

Keywords Data

Google Ads search volume, CPC, competition and a 12-month trend for up to 1,000 keywords per call.

$0.09 per call

DataForSEO Labs

Keyword suggestions, what any domain ranks for, organic competitors, keyword gaps and traffic estimates.

$0.012 a task + $0.00012 a row

Backlinks

Referring domains, anchors, new and lost links for any domain or URL.

About $0.024 a request + rows

On-Page API

Crawl and check pages through the API: tags, status codes, page speed and on-page issues.

Pay per page crawled

AI Optimization

LLM Mentions, LLM Responses and LLM Scraper for ChatGPT, Gemini, Claude, Perplexity and AI Overviews.

LLM Mentions $0.10 a request + $0.001 a row

Prices from DataForSEO pricing, its SERP and AI Optimization pages. Since July 2026, Backlinks and LLM Mentions are pay-as-you-go like everything else; the old $100 monthly commitment is gone, and Backlinks rates rose 20%.

How the pieces fit together

You set the outcome and the rules. Claude pulls the data and hands back something you can ship.

YouThe outcome you want+ your rules(cite it, report cost)ClaudeClaude Code or Coworkreads your skill file,writes code, calls APIsDATA IT PULLSDataForSEO APISite crawler exportSearch Console MCPGA4 MCPYour CMS or repoDeliverableCSV, report, fix list,new page or redirect file+ what the API calls costYou review and ship. Anything that worked gets saved to the skill file for next time.
Diagram: Claude connected to DataForSEO, a crawler export and your analytics.

MCP, short for Model Context Protocol, is one way to connect the assistant to those tools. A skill is a different piece: saved instructions explaining how to do a recurring job. The connection gives access; the skill holds your process. You can use a connection without a custom skill, then save the workflow once you know it works.

You can see the same pattern across the industry. SE Ranking offers Claude skills that turn connected data into briefs and audits, and Ahrefs writes about encoding editorial expertise into skills. DataForSEO is the data layer in this guide, and its flexibility makes it a good base for building your own versions.

SEO teams already lean on Claude, but only 26% connect a data API

Tools SEO teams use for AI work, up to three per respondent. The data connection is the part most teams skip.

ClaudeChatGPTOther AI assistantsSEO data APIsAutomation and code
ClaudeClaude: 78%78%ChatGPTChatGPT: 57%57%GeminiGemini: 33%33%SEO data APIsSEO data APIs: 26%26%PerplexityPerplexity: 15%15%Internal AI workflowsInternal AI workflows: 15%15%Zapier / MakeZapier / Make: 10%10%Custom scripts / PythonCustom scripts / Python: 7%7%
Source: Keyword.com, State of AI and Automation in SEO 2026, n=96.
Make the connection

Connect DataForSEO and prove it works

Four steps, about ten minutes, and a test that costs pennies.

Diagram of the Claude app connected through an MCP plug to the DataForSEO API hub, which fans out to SERP, Keywords, Labs, Backlinks, On-Page and AI Optimization endpoints

One connection, six data sources. Claude talks to DataForSEO through MCP; your credentials stay in the connection settings.
1Open the accountsClaude Pro or higher for Claude Code. A DataForSEO account with the $50 minimum top-up.
2ConnectAdd DataForSEO's MCP server, or a skill file that loads your API login from a local .env file.
3Add your siteDrop a crawl export, Search Console data and a short business description into a project folder.
4Run a tiny testFive keywords and one live SERP. Check the raw data and the cost before you scale up.

Claude Pro is $20 a month billed monthly and includes Claude Code, with usage limits. DataForSEO is pay-as-you-go, with a $50 minimum top-up and a balance that doesn't expire. That top-up is credit you spend on requests, not a recurring charge.

Follow DataForSEO's current instructions for your client. The hosted MCP endpoint is https://mcp.dataforseo.com/v3/mcp, and there's a local option that runs with npx -y dataforseo-mcp-server. Keep credentials in the connection settings or a local environment file, away from prompts and shared files.

Then run a deliberately small test. Ask for US search volume for five terms and one live Google results page, and require the raw data and the returned request costs. It proves the connection works and gives you a baseline for what a real result looks like.

Your first testabout 10 cents
Use the connected DataForSEO tools to retrieve US English search volume for these five keywords, then fetch the live desktop Google results for the first one. Save the responses and a short summary. State the volume period, endpoint, retrieval date and API cost. Stop if the connection fails; leave unavailable metrics blank.
You getProof the connection works, plus a cost baseline

Which crawler do you need?

Put your crawl and analytics exports in a project folder with a short description of the business: what it sells, where it operates, which pages matter and what counts as a conversion. That context helps Claude tell a commercially useful opportunity from a big keyword that just happens to be related.

Screaming Frog is free up to 500 URLs and $279 a year after that. Sitebulb starts at $18 a month. Claude can crawl a small site itself, but I wouldn't go past about 10 pages, and it won't be as thorough as a real crawler.

Claude can crawl about 10 pages. A real crawler handles the rest

How many pages each option handles comfortably, log scale. Arrows mean no fixed cap.

Claude aloneScreaming FrogSitebulb
1101001K10K100K+Claude aloneIncluded with your planClaude alone: Up to ~10 pages, spot checks onlyUp to ~10 pages, spot checks onlyScreaming Frog, free$0Screaming Frog, free: 500-URL cap500-URL capScreaming Frog, paid$279 a yearScreaming Frog, paid: No URL cap; limited by your machine's memoryNo URL cap; limited by your machine's memorySitebulb Lite / Pro$18 / $42 a monthSitebulb Lite / Pro: Full-site audits with JavaScript crawlingFull-site audits with JavaScript crawling
Sources: Screaming Frog pricing, Sitebulb pricing via Capterra. The 10-page ceiling is my own rule of thumb.
A Claude prompt flows through a DataForSEO funnel into a finished results table, with a stopwatch reading 30 seconds and a two-cent coin

Two real runs

Two small jobs show why the setup is useful

Here are two jobs I ran on October 15, 2026, with the exact prompt, what came back, how long it took and what the API charged.

Run 1 · Striking distance

Find the page that's one push from page one

30 secto pull and sort the data
$0.021actual API cost
2API calls

Input

Find every keyword vuss.io ranks 8 to 20 for in the US. Group them by page, check the live SERP for the biggest term, and tell me what to change.
Calls Claude made
  • ranked_keywords positions 8 to 20$0.019
  • serp/google/organic live top 10$0.002
Result

59 keywords sit in positions 8 to 20, and every one of them points at a single page: the high-resolution Where's Waldo images post.

17 at 8 to 10
26 at 11 to 15
16 at 16 to 20
  • The biggest variations, like "where's wally images" (#8) and "wheres wally pictures" (#9), show 8,100 searches a month each. Google groups close variants, so those overlap and shouldn't be added up.
  • The title never says "pictures" or "Wally," the two words most of these searches use.
  • The live top 10 is all galleries and image libraries (Shutterstock, Pinterest, Candlewick), so the page's format already fits.
Current titleMind-Blowing High-Resolution Where's Waldo Images - Vussio
Claude's rewriteWhere's Waldo Pictures: High-Res Images (Where's Wally)

Plus two new sections the data asked for: "hardest Where's Waldo" (ranking 13 to 19) and the beach scene (10 to 17).

Before you ship it
  • Make sure the new title still describes the page accurately.
  • Treat the ranking data as a lead, not proof that a missing word caused the position.
  • On your own site, confirm with Search Console. A third-party ranking snapshot and Search Console's average position measure different things.
  • Change it, note the date, and measure over a consistent period.
Run 2 · Keyword research

Plan a drone photography site from one seed

22 secto pull and cluster 200 keywords
$0.036actual API cost
1API call

Input

Pull every "drone photography" keyword with 30+ US searches a month, group them into the pages a drone photography business should build, and tell me which to build first.
Calls Claude made
  • keyword_suggestions 200 rows$0.036

Lesson from the first try: I started with keyword_ideas, which drifted into "pilot flying j" and webcams ($0.048 wasted). keyword_suggestions stays close to the seed, and its exact_match option keeps the literal phrase.

Result: 200 keywords, 9 pages
Page to buildKwsLead termKD
Real estate drone photography build first22real estate drone photography · 1,9000
Local service page ("near me") build first9drone photography near me · 1,6002
Pricing and rates build first22drone photography cost · 2600
Core service page21drone photography · 4,40015
Best drones for photography (guide)66best drone for photography · 1,30021
Wedding drone photography10drone wedding photography · 3200
Learn drone photography18drone photography tips · 2103
Commercial and construction10commercial drone photography · 1104
City pages (only your own city)22drone photography houston · 900

KD is the median keyword difficulty for the cluster (0 to 100). The real estate and "near me" pages target 1,900 and 1,600 monthly searches with almost no competition, so they go first.

Before you ship it
  • Check each proposed page against what the business actually sells. A drone-buying guide may attract people who will never hire a photographer.
  • Build city pages only where the business operates.
  • A low difficulty score screens ideas; it doesn't prove a local query is easy. Look at the live results first.
  • Pull a small sample before a large request. It saves cleanup and API spend.

Times are wall-clock from the first API request to the finished table; add a few minutes to read it over. These runs show research output and request costs, not measured ranking gains.

Free tool · 16 jobsRun these same jobs on your own siteType in your domain once. Every prompt rewrites itself, ready to paste into Claude.Open the SEO Prompt Builder

Swiss Army knife with a Claude and DataForSEO handle and sixteen color-coded SEO tools fanned out
Interactive toolFreeNo sign-up16 SEO jobs

The SEO Prompt Builder

Type in your site once and all 16 prompts rewrite themselves for you. Pick a job, copy the prompt, paste it into Claude. That's it.

  1. 1
    Enter your detailsWebsite, market, topic and two competitors.
  2. 2
    Pick a jobFilter by category below.
  3. 3
    Copy and paste into ClaudeClaude pulls the DataForSEO data and does the work.
1Enter your detailsOptional. Anything you leave blank stays as a [bracket] to fill in later.
2Pick a job
3Copy a prompt and paste it into ClaudeShowing all 16 jobs

Keyword research

Keyword research and clustering

Turn a seed topic into a short list of pages worth building.

BringA topic, your website and a description of what the business sells.
Ask Claude~22 sec
Research [TOPIC] for [DOMAIN], targeting [MARKET]. Start with a small relevant sample, then pull up to 200 keyword suggestions and available volumes. Group by intent, check representative live SERPs, and map each group to an existing or proposed page. Explain what to build first and which terms to exclude. Export a keyword CSV and a prioritized page map. Report sources, limits and API cost; do not sum close-variant volumes.
You getA keyword CSV and a prioritized page map.

See example outputsample data
Keyword research page map: intent groups mapped to existing and proposed pages, with exclusions and API cost

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Striking-distance opportunities

Find existing pages where a focused edit is worth testing.

BringA Search Console export with page, query, country, device and dates.
Ask Claude~30 sec
Use my Search Console export for [DOMAIN] in [MARKET]. Find page-query pairs with average position 8 to 20 and meaningful impressions during the last complete 28 days. Compare with the preceding 28 days, group by page, and check a representative live SERP for the strongest candidates. Recommend five pages to investigate. Draft specific title or content edits and explain the evidence. Keep observed changes separate from hypotheses. Save a CSV and report API cost.
You getFive page recommendations with draft edits and evidence.

See example outputsample data
Search Console striking-distance report: five pages in positions 8 to 20 with draft title and content edits

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Data: ~$0.02 data (Run 1)Walkthrough

Competitor keyword gap

Find relevant gaps without copying a competitor's whole strategy.

BringYour domain and two genuine search competitors.
Ask Claude3 calls
Compare [DOMAIN] with [COMPETITOR 1] and [COMPETITOR 2] in [MARKET], focusing on [TOPIC]. Find relevant queries the competitors rank for that our returned ranking data does not show. Filter out unrelated services, locations and branded navigation. Inspect the live results for the best candidates. Give me the top 20 opportunities, the page to update or create, and the reason each belongs in our plan. State coverage, filters, sources and API cost.
You getA ranked opportunity list mapped to pages.

See example outputsample data
Competitor keyword gap: filter funnel and top opportunities mapped to pages to update or create

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Data: ~$0.27 dataWalkthrough

Technical SEO

Crawl audit to fix list

Turn hundreds of crawler warnings into work someone can implement.

BringA crawl export, important templates and available performance data.
Ask Claude1 crawl export
Analyze this crawl of [DOMAIN] alongside the supplied Search Console or analytics data for [MARKET]. Group issues by template and likely cause. Create the ten most useful implementation tickets, each with affected URLs, evidence, a proposed fix, priority, confidence and a verification step. Distinguish observed problems from hypotheses. Where performance data is unavailable, prioritize by technical impact and business importance rather than invented traffic estimates. Export the tickets and supporting URL list.
You getTen implementation tickets with checks for completion.

See example outputsample data
Crawl audit turned into ten implementation tickets with evidence, priority, confidence and verification steps

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Migration redirect map

Match old URLs to useful destinations and isolate the uncertain cases.

BringOld and new crawls with titles and main content.
Ask Claude2 crawl exports
Build a redirect proposal for [DOMAIN] using these old and new crawl exports. Match URLs using paths, titles, main content and visitor intent. Include a reason and confidence for every match. Leave genuinely unmatched or uncertain URLs for review. Check destinations for status, indexability, redirect chains and loops. Export a redirect CSV and an exceptions file. Prepare the proposal for review without applying changes.
You getA redirect CSV and a separate exceptions file.

See example outputsample data
Migration redirect map: matched URLs with reasons and confidence, destination checks and an exceptions file

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Structured-data review

Generate markup that fits the actual page and current requirements.

BringRepresentative page content and existing structured data.
Ask Claudeno API spend
Review the visible content and existing structured data on these [DOMAIN] templates. Check current official requirements before recommending types. Identify missing, conflicting or unsupported markup, then propose only the JSON-LD supported by the actual page content. Validate syntax and distinguish valid markup from rich-result eligibility. Do not invent product, review, organization or author facts. Deliver a report and proposed changes for review, without publishing them.
You getA validation report and proposed JSON-LD changes.

See example outputsample data
Structured-data review by template: issues, proposed JSON-LD, validity and rich-result eligibility

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

AI crawler log check

See which sections bots requested and what responses they got.

BringA server-log sample and your robots.txt or crawler policies.
Ask Clauderuns locally
Write and run a parser for these [DOMAIN] server logs. Separate training crawlers, search crawlers and user-directed fetch agents using current official documentation. Summarize requests by agent, site section, date and status code. Flag blocked requests and unverified user-agent-only identifications. Compare with the supplied crawler policies. Do not infer citations, referrals or revenue from crawl hits. Save the parser, summary tables and a short investigation list.
You getA reusable parser and a report by bot, section and status.

See example outputsample data
AI crawler log check: requests by bot type, status codes, IP verification and policy mismatches

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Content + AI visibility

Brief from the live SERP

Give a writer a clear angle and the evidence to back it.

BringA topic, target audience, source material and URL inventory.
Ask Claudelive SERP
Build a content brief for [TOPIC] for [DOMAIN], targeting [MARKET]. Use current live results and available keyword data to determine intent and page format. Combine that research with the supplied expertise, interviews or original examples. Recommend an original angle, section outline, required evidence and internal links from our URL inventory. Flag claims needing an expert or more research. Save a writer-ready brief with source links and API cost.
You getA writer-ready brief with sources and internal links.

See example outputsample data
Content brief built from the live SERP: angle, section outline, internal links and claims flagged for an expert

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Bulk titles and descriptions

Prepare metadata edits that keep each page's purpose intact.

BringA crawl export and the top queries for each page.
Ask Claude20 pages
Join this [DOMAIN] crawl export to the supplied Search Console page-query data for [MARKET]. Propose titles and meta descriptions for the 20 pages with the clearest need. Keep wording accurate, distinct and matched to page intent. Aim for concise display lengths without treating character counts as guaranteed Google limits. Preserve brand and product facts. Export current text, proposed text, target query and reason in a CSV. Leave uncertain pages for review.
You getA before-and-after CSV with a reason for each change.

See example outputsample data
Bulk titles and meta descriptions: before-and-after rows with target query and reason for each change

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

AI citation tracking

Track a defined sample without confusing mentions with citations.

BringTen customer questions, a platform and a fixed measurement method.
Ask Claude10 questions
For [DOMAIN] and [TOPIC] in [MARKET], propose ten useful customer questions for AI visibility checks. Specify a supported DataForSEO measurement method before collecting data, then keep that method consistent. Record platform, model or interface, location, date, answer and cited URLs. Distinguish brand mentions, retrieved search results and actual citations. Save the results and a repeatable tracking plan. State sample limits and API cost; do not present one run as total market visibility.
You getAn answer-and-citation log ready for repeat checks.

See example outputsample data
AI citation log: ten questions with mentions, retrieved results and citations counted separately

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Original research study

Investigate a useful question with reproducible evidence.

BringA hypothesis and a defined list or sample of public sites.
Ask Claudereproducible
Help design a small original study about [TOPIC] for [DOMAIN]. Define the hypothesis, sample, measurements and limits before collecting data. Use supplied sites or a documented selection process, with DataForSEO and direct public-page checks where appropriate. Save raw observations, sources and retrieval dates, then produce tables and a draft methods-and-findings summary. Separate observation from causation. Treat missing search results as inconclusive evidence of indexing, and report API cost.
You getA methods note, source table and supported findings.

See example outputsample data
Original research study: hypothesis, sample accounting, citations by domain type and observed findings

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Reporting

Weekly rank tracking

Produce a consistent snapshot and a meaningful change report.

BringA fixed keyword list, target URLs, market and device.
Ask Claude$0.24/mo
For this fixed keyword list and [DOMAIN], retrieve Google ranking results for [MARKET] using a consistent device and depth. Save the date, query, returned URL, position and relevant SERP features. Compare with the supplied previous snapshot and flag changes needing investigation. Distinguish not found within the requested depth from absent everywhere. Report API cost. Save a reusable tracking specification; do not claim weekly monitoring is active unless a scheduler is actually configured.
You getA ranking snapshot, comparison and repeatable tracking spec.

See example outputsample data
Weekly rank tracking snapshot: position changes, SERP features and a saved tracking spec

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Data: $0.24/mo for 100 keywords queuedWalkthrough

GA4 report on demand

Connect organic landing-page activity to real business outcomes.

BringGA4 access or an export, date ranges and an agreed conversion definition.
Ask ClaudeGA4 MCP
Using connected GA4 data or the supplied export for [DOMAIN], compare the last complete 28 days with the preceding 28 days. Analyze organic landing pages, sessions and the business's defined key events. Keep property, channel, timezone, event definitions and market filters consistent. Identify where changes are concentrated, flag data-quality limits, and separate facts from possible causes. Save the underlying tables and a concise report. State the data source and exact periods.
You getA concise report plus the underlying landing-page tables.

See example outputsample data
GA4 organic landing page report: 28-day comparison, key events, facts separated from possible causes

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Paid and organic overlap

Find candidates for an incrementality test.

BringAligned Search Console and Google Ads query data; analytics if available.
Ask ClaudeGSC + Ads
Join the supplied Search Console and Google Ads search-term data for [DOMAIN] in [MARKET], using compatible dates and dimensions. Identify queries with paid clicks and strong organic visibility. Separate brand and non-brand queries, report conversion evidence where available, and flag unmatched or incomplete records. Produce a shortlist for incrementality testing and a proposed test design. Do not label spend wasted or recommend pausing ads solely because an organic result ranks highly.
You getAn overlap shortlist and a test proposal.

See example outputsample data
Paid and organic overlap: non-brand test shortlist and a proposed incrementality test design

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Ecommerce + local

Product listing audit

Improve listing relevance while keeping product details accurate.

BringA catalog export, product facts and the target sales channel.
Ask Claude20 listings
Review these [DOMAIN] product listings for [TOPIC] in [MARKET]. Preserve product specifications and factual claims. Use relevant keyword evidence to identify mismatched or unclear titles and descriptions, then propose edits for the 20 strongest candidates. Distinguish Google search demand from marketplace-specific behavior; do not assume they are equivalent. Export current wording, proposed wording, evidence and confidence. Flag missing product facts rather than inventing them, and report API cost.
You getA prioritized listing review with proposed titles or descriptions.

See example outputsample data
Product listing audit: proposed titles with marketplace and Google evidence, missing product facts flagged

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Local SERP check

See what people in the actual service area are likely to find.

BringA service, a specific location and your service area.
Ask Claude$0.002/SERP
Investigate [TOPIC] for [DOMAIN] in [MARKET] using a specific supported search location and consistent device. Retrieve live organic results and local-pack information where supported. Separate directory pages, service providers and informational pages. Compare the actual service area and supplied business information with the result types. Recommend three relevant page improvements supported by the evidence. Report location, retrieval date, coverage and API cost; avoid treating a national SERP as a local result.
You getA local competitor snapshot and three page improvements.

See example outputsample data
Local SERP check: local pack, result types by query and three page improvements

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

Data: $0.002 per live SERPWalkthrough
The walkthroughs

Six workflows, and the decisions that make them work

The prompts are a starting point. These are the judgment calls that turn the output into useful SEO work.

01

Keyword research that ends with a publishing decision

Start by describing the customer and what the business actually offers. Have Claude expand a focused seed list, pull metrics in batches and suggest clusters. Then have it check representative live results before assigning each cluster to a page. Similar wording can hide different needs: someone researching camera gear wants something different from someone hiring a photographer.

The deliverable should show what to publish, what to update and what to leave alone. For each recommendation, ask for a URL, the intended reader, the page format, a lead query and the reasoning. Include an "exclude" group so irrelevant keywords don't quietly become assignments. A useful extra check: ask which clusters compete for the same results. That reveals groups that belong on one page.

Try this prompt
Research US English keywords for this business. Propose up to ten page groups, check a representative live SERP for each, and map them to our existing URLs where possible. Deliver a keyword CSV and a prioritized page plan. Explain exclusions and uncertain groupings. Keep close-variant volumes separate from any estimate of total demand.
You getA keyword CSV and a prioritized page plan

See example outputsample data
Keyword research page map: intent groups mapped to existing and proposed pages, with exclusions and API cost

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

02

A crawl audit that produces implementation tickets

A crawl can produce hundreds of warnings without telling you which five fixes matter this week. Give Claude the crawl, a description of the important templates and any Search Console or conversion data. Ask it to group repeated problems by likely cause. One template fix can clear a hundred individual warnings.

The output gets useful when it includes the actual URLs, evidence, the proposed change and a way to verify it. A missing meta description needs a different response than an accidental noindex on a money page. Where business data is missing, Claude should say so instead of inventing "traffic at risk." And a link crawl can't find every orphan page alone: supply a sitemap or CMS inventory to compare against.

Try this prompt
Join this crawl to the supplied page-performance data. Group issues by template and likely cause. Create the ten most useful implementation tickets, each with affected URLs, evidence, a proposed fix, confidence and a verification step. Separate observed problems from hypotheses and explain the priority order.
You getTen implementation tickets you can hand to a developer

See example outputsample data
Crawl audit turned into ten implementation tickets with evidence, priority, confidence and verification steps

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

03

Content briefs that give a writer an angle

Claude can read live results, find recurring questions and outline a page. The brief gets far more valuable when you add material competitors don't have: an interview, product documentation, customer questions, original measurements or a worked example. Otherwise you get an organized version of what readers have already seen.

DataForSEO documents a practitioner workflow that moves from customer information through keyword and SERP research to a page concept and brief. The lesson that transfers: let the research shape the format and angle before anyone starts drafting. For an existing article, also ask what to cut or combine.

Try this prompt
Build a brief from the live results and these source materials. Explain the reader's task, the right page format, the original contribution we can make, and the evidence each section needs. Suggest internal links from our URL inventory. Flag claims that need an expert, a source or a new test.
You getA writer-ready brief with sources and internal links

See example outputsample data
Content brief built from the live SERP: angle, section outline, internal links and claims flagged for an expert

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

04

Internal links and redirect maps with reasons attached

Give Claude a crawl with page titles, extracted main content and existing links. It can find related pages that barely connect and propose links in specific passages: source page, destination, anchor and the surrounding sentence. That's an edit someone can review and ship.

Screaming Frog's semantic-similarity analysis is a useful shortcut for surfacing related pages and candidate redirect matches. Extract the main content first, since repeated menus and footers make different pages look alike. Similarity finds candidates; intent and content decide the final match.

Try this prompt
Using these old and new crawls, propose a redirect map with a reason and confidence for every match. Leave uncertain or genuinely unmatched URLs for review. Check proposed destinations for status, indexability, chains, loops and relevance. Export the map and a separate exceptions file.
You getA redirect map plus an exceptions file

See example outputsample data
Migration redirect map: matched URLs with reasons and confidence, destination checks and an exceptions file

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size

05

Reporting that keeps a change separate from its cause

Reporting is a strong recurring job because the inputs and output can be defined once. Give Claude consistent date ranges, a conversion definition and the right account data. Google provides an official Analytics MCP server, but you can start with exports and keep version one simple.

Ask Claude to find where a change is concentrated, then bring in the evidence to investigate it. Falling clicks on one template deserve a different investigation than a sitewide dip. Remember that Search Console's query API doesn't guarantee every row, so document filters and limits before calling a dataset complete.

Try this prompt
Compare the last complete 28 days with the preceding 28 days for organic landing pages. Keep country, device, channel and conversion definitions consistent. Identify the largest changes, check data completeness, and list observed facts separately from possible causes. Deliver a short report and the underlying tables.
You getA short report with the underlying tables

See example outputs (2)sample data
GA4 organic landing page report: 28-day comparison, key events, facts separated from possible causesGA4 and Search Console traffic check: organic sessions reconciled with clicks and a bot spike flagged

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size 1 · Open full size 2

06

AI visibility checks with a defined measurement method

AI search research is useful when you know what the result represents. DataForSEO's AI Optimization suite separates several products: LLM Mentions searches a database covering ChatGPT and Google AI Overviews, LLM Responses queries models directly, and LLM Scraper captures consumer interfaces. They answer different questions, so report them separately.

A result retrieved during a web search isn't necessarily a source cited in the answer. DataForSEO's citation guide separates those fields. For bot logs, keep training crawlers apart from search and user-request agents; Anthropic documents separate roles for ClaudeBot, Claude-SearchBot and Claude-User. A crawl hit doesn't prove a citation or a visit.

Try this prompt
For these ten customer questions, use one clearly specified measurement method. Save the platform, model or interface, location, date, answer and cited URLs. Distinguish brand mentions from citations and retrieved search results. Repeat the same method over time and report changes within this sample.
You getA repeatable answer-and-citation log

See example outputs (2)sample data
AI citation log: ten questions with mentions, retrieved results and citations counted separatelyAI crawler log check: requests by bot type, status codes, IP verification and policy mismatches

Illustrative mockup of what Claude hands back for vuss.io, built with sample data. Open full size 1 · Open full size 2

ClaudeBot more than doubled its share of AI crawler traffic, from 9% to 20%

Share of AI crawler traffic, Cloudflare Radar, May 2026.

GooglebotGooglebot: 27%27%ClaudeBotClaudeBot: 20%20%was 9%Meta-ExternalAgentMeta-ExternalAgent: 12%12%GPTBotGPTBot: 10%10%
Source: Cloudflare Radar figures cited by Similarweb. Shares shift month to month.

Free tool · copy and pasteAll six walkthroughs, pre-filled for your sitePick a job in the SEO Prompt Builder, copy it, and let Claude do the pulling and sorting.Open the SEO Prompt Builder

Make it repeatable

The small habits that make this setup useful

Be precise about what should come back, and save what works.

  1. Ask for the deliverable, not advice. "An SEO audit" leaves a lot for Claude to decide. "Ten implementation tickets with affected URLs and verification steps" is a job you can assess.
  2. Set the frame up front. Market, date range, files, business goal, output format and a spending limit.
  3. Save raw data before analysis. Pull a keyword dataset once and reuse it for clustering, the brief and the report. Every surprising recommendation stays traceable.
  4. Let code do the math. Joins, dedupes, calculations and validation run as scripts; the language model handles interpretation.
  5. Trim what you feed it. Cutting a Screaming Frog export from about 70 columns to 15 saves 52% of the tokens. More in my Claude cost breakdown.
  6. Save what worked as a skill. Include the decisions that mattered: exclusions, required evidence, how uncertainty should appear and what needs review. Anthropic's own marketing team works this way.

Keep your instructions current

"Generate FAQ schema" used to be on every SEO task list, but Google stopped showing FAQ rich results on May 7, 2026. For Google's AI features, its documentation says no special AI markup is required. A good workflow checks today's requirements before generating another file.

The payoff

Where the time comes back

The savings come from pulling, sorting and formatting: the part of SEO that used to fill whole afternoons.

In Keyword.com's 2026 survey, 89% of SEOs using AI save at least 4 hours a week, and a third save more than 10. How much you get back depends on how much of your week is that kind of work.

89% of SEOs using AI save at least 4 hours a week

Teams with AI central to delivery were far more likely to save 7+ hours (81%) than teams still testing it (11%).

1 to 3 hrs: 11% of respondents11%1 to 3 hrs4 to 6 hrs: 36% of respondents36%4 to 6 hrs7 to 10 hrs: 21% of respondents21%7 to 10 hrs10+ hrs: 33% of respondents33%10+ hrs89% save 4+ hours every weekHours saved per week, share of SEO teams using AI (n=92). Figures sum to 101% due to rounding.
Source: Keyword.com, 2026. The report notes this is an association, not proof of cause.

The same jobs, manual vs with an agent

Time per task, log scale. Some are vendor claims, so treat them as a direction.

ManualWith an agent
1 min10 min1 hr8 hrsWriter-ready content briefSE Ranking workflow (vendor claim)Writer-ready content brief, manual: 3 to 4 hrsWriter-ready content brief, with an agent: <10 min3 to 4 hrs<10 minAI search visibility reportSE Ranking workflow (vendor claim)AI search visibility report, manual: a full dayAI search visibility report, with an agent: a few mina full daya few minBuyer-intent keyword mappingPractitioner, via Discovered LabsBuyer-intent keyword mapping, manual: a full dayBuyer-intent keyword mapping, with an agent: ~20 mina full day~20 minCustomer case study draftAnthropic marketing teamCustomer case study draft, manual: 2.5 hrsCustomer case study draft, with an agent: 30 min2.5 hrs30 min
Sources: SE Ranking, Discovered Labs, Anthropic. "3 to 4 hrs" plotted at 3.5; "a full day" at 8 hrs; "a few min" at 5.
The economics

What it costs, and where the value comes from

For focused jobs, the data costs cents. Most of the cost is the subscription and your review time.

Balance scale weighing a tall stack of monthly SEO platform invoices against a few coins plus Claude and DataForSEO, with a 9-cent receipt

The two runs above used $0.021 and $0.036 of DataForSEO requests. The keyword volumes behind this article (35 terms in two calls) cost $0.18. Those numbers cover search data only, not the AI plan, a crawler license or the time to check the work. Plan for all four.

Here's the same work done both ways. Data costs are DataForSEO's published rates; Claude's work runs inside your subscription.

The jobIn SemrushWith Claude + DataForSEOData cost
Volumes for 1,000 keywords
SemrushKeyword Overview / Keyword Magic Tool on a paid plan
Claude + DataForSEOOne search_volume call. Claude dedupes, clusters and exports the CSV.
$0.09
Keyword gap vs 3 competitors
SemrushKeyword Gap tool
Claude + DataForSEOThree domain_intersection calls at 1,000 rows each, sorted into a target list.
$0.40
A competitor's top keywords
SemrushOrganic Research
Claude + DataForSEOranked_keywords for the domain, 1,000 rows, grouped by page.
$0.13
Track 100 keywords weekly
SemrushPosition Tracking project
Claude + DataForSEO100 standard-queue SERPs at $0.0006 each every Monday, with a change report. Live mode costs $0.002 a SERP.
$0.24 a month ($0.80 live)
Full site audit
SemrushSite Audit (crawl limits set by plan)
Claude + DataForSEOA Sitebulb or Screaming Frog export, prioritized by Claude against ranking data.
$0 data, plus your crawler
Backlink profile
SemrushBacklink Analytics
Claude + DataForSEObacklinks/summary plus referring domains, summarized into wins and risks.
~$0.06
AI answer visibility
SemrushAI Visibility (Semrush One, from $199/mo)
Claude + DataForSEOllm_mentions for 10 money questions, with who gets cited and why.
~$1

The everyday setup runs $540 to $756 a year. Semrush Guru is $2,999

Annual cost in USD at list prices.

Claude + DataForSEO + a crawlerSemrush plans
Claude Pro + DataForSEO + Screaming Frog (free)Claude Pro + DataForSEO + Screaming Frog (free): $540$540Claude Pro + DataForSEO + Sitebulb LiteClaude Pro + DataForSEO + Sitebulb Lite: $756$756Semrush ProSemrush Pro: $1,679$1,679Claude Max + DataForSEO + Sitebulb ProClaude Max + DataForSEO + Sitebulb Pro: $2,004$2,004Semrush GuruSemrush Guru: $2,999$2,999Semrush Business (API access, units extra)Semrush Business (API access, units extra): $5,999$5,999
Semrush list prices from this 2026 breakdown; Semrush API units are sold separately (SE Ranking). Claude Pro $20/mo, Claude Max $100/mo, Sitebulb $18 or $42/mo. DataForSEO assumes about $25/mo of calls.

Whether this replaces a paid SEO platform depends on the work you need. Teams may still value an established interface, saved projects, historical data and ready-made monitoring. This setup shines when your question crosses tools, or when the output you need doesn't exist in a standard dashboard.

Capability Claude + DataForSEO Semrush
Volumes, SERPs, rankings, competitors, gaps Yes Yes
Backlinks and AI answer mentions Yes (pay as you go) Yes (higher tiers)
Traffic estimates Estimated from rankings Yes (clickstream)
Deep history and a polished UI You build what you need Yes
Writes the report, fix or page for you Yes No

Does Semrush get its data from DataForSEO?

I've heard the claim and couldn't confirm it, so I won't state it as fact. Semrush's data page says it partners with unnamed third-party providers for Google SERPs, uses a clickstream panel for traffic and its own crawler for backlinks. What is public: plenty of SEO tools openly run on DataForSEO, like SEO Utils.

Claude is the example throughout, but the approach works in other assistants that support tools, files and code execution. OpenAI documents MCP support for Codex. Use the environment you're comfortable with and confirm it can reach the data your task needs.

Your first job

Start with one page and one decision.

Pick a page you already care about. Supply its content, its Search Console data and a crawl if you have one. Have Claude check the live results through DataForSEO and prepare three changes, each backed by evidence. Set a small spending limit, review the output and ship the strongest one.

Start with this jobcapped at $0.25
Investigate this important page on [DOMAIN] for [TOPIC] in [MARKET]. Use the supplied page content, Search Console export and crawl, then check a representative live SERP through DataForSEO. Prepare three specific improvements with evidence, proposed wording or implementation steps, confidence and a verification plan. Save a short report and source tables. Cap DataForSEO spend at $0.25. Report actual API cost and leave missing metrics blank. Prepare the changes for review without publishing them.
You getThree evidence-backed fixes, ready for review

Record the change and check performance over a sensible period. When the workflow earns its place, save it as a skill. Then add a second job: a page plan, a fix list, an internal-link review or a recurring report. That's how the knife gets useful: learn what each tool does, pick the one the job needs, and finish with something you can ship.

Browse all 16 jobs in the SEO Prompt Builder

More Writing & Research by Jourdan Rombough

View all research

SEO MEDITATIONS