# How to do programmatic SEO on a new domain: a 30-day plan

> A 30-day programmatic SEO plan for a brand-new site: Grok for research, real volume checks, verified data, small launch batches and a day-30 indexation review.

Maciej Wiatr, 2026-10-31. https://maciejwiatr.com/blog/how-to-do-programmatic-seo-new-domain

Programmatic SEO on a new domain is a different job from the Wise and Zapier case studies most guides use. Those sites had authority before they had templates. If you are starting from zero, the steps are the same: find a pattern, get the data, build a template and publish. The scale has to be smaller and the checks stricter. This post is a 30-day plan for a site with no SEO history. It uses Grok (the xAI Responses API with `web_search` and `x_search`) as a research assistant, not as a page writer. It also covers when a new site should not start at all.

## Does programmatic SEO still work?

Yes, when each page carries data the next page doesn't. Ahrefs' [programmatic SEO guide](https://ahrefs.com/blog/programmatic-seo/) shows QuillBot getting about 464K visits from about 130 pages, and Canva growing from 23 to about 310 pages. A [512-page B2B SaaS case](https://thestacc.com/blog/programmatic-seo-case-study/) went from 240 to 11,840 monthly clicks over 18 months. Every page in that case had at least 3 unique data points.

The sites that lost in 2026 had scaled pages that added little. In [Glenn Gabe's August 2026 spam update cases](https://gsqi.com/marketing-blog/august-2026-google-spam-update-case-studies/):

- a programmatic plus AI site in a YMYL niche lost 200K+ queries
- a site with 1.5M indexed URLs, 85% of them programmatic, was hit site-wide

Gabe's question fits any new build: "If anyone can set up a site programmatically like this with almost no human involvement at all, what value is there?"

**Is SEO dead now with AI?** No. [Backlinko](https://backlinko.com/programmatic-seo) blames part of recent traffic losses on AI Overviews. Gabe notes that spam updates also cut visibility in AI Overviews and AI Mode, so the same quality rules decide whether you show up in either place. I covered the 2026 updates in [Is programmatic SEO dead?](https://maciejwiatr.com/blog/is-programmatic-seo-dead-google-updates).

## When a new site should not do programmatic SEO yet

Backlinko says pSEO is "probably not right for most sites, at least not yet" if you lack authority, resources or unique data. [GTM 80/20](https://www.gtm8020.com/blog/early-stage-programmatic-seo.md) is more specific. It says to skip pSEO if any of these is true:

- your patterns don't reach roughly 100+ variants and about 10K monthly searches
- you have no proprietary data
- you haven't written 20 to 30 pieces by hand
- you can't commit 6 to 12 months

I'd treat the data condition as the hard one. If the only thing that changes between pages is a keyword, Google's [spam policies](https://developers.google.com/search/docs/essentials/spam-policies) describe your site under doorway abuse: pages "created to rank for specific, similar search queries."

## How to do programmatic SEO from zero: the 30-day plan

Ahrefs and Backlinko agree on the core steps:

1. Find a repeating pattern where search intent stays the same across variations.
2. Study the top results to design the template.
3. Get value-added data.
4. Test a few pages by hand.
5. Publish in stages.
6. Monitor indexing and prune.

The plan below fits those steps into 30 days and shows where Grok helps and where it stops.

```mermaid
flowchart TD
    A[Grok finds question patterns] --> B[Check volume elsewhere]
    B --> C[Build and verify dataset]
    C --> D[Hand-test 3-5 pages]
    D --> E[Launch first batch]
    E --> F[Read indexation report]
```

### Days 1 to 7: How can Grok help with SEO research for a brand-new site?

A new site has no Search Console data to mine, so you need demand signals from somewhere else. Grok's search tools give you three:

- **`x_search`** finds recurring questions in recent posts, such as "X vs Y" or "best X for Y". You can set `from_date` and `to_date` in YYYY-MM-DD format, and `to_date` is exclusive. You can limit results to up to 20 niche handles with `allowed_x_handles`. `max_turns` caps how many tool rounds the model runs ([xAI X Search docs](https://docs.x.ai/developers/tools/x-search)).
- **`web_search`** with `allowed_domains`, capped at 5 domains, lets you inspect competitor templates and which page types rank ([xAI search tools docs](https://docs.x.ai/docs/guides/tools/search-tools)).
- **Citations** come back with each response, so you can open every source yourself.

The API uses base URL `https://api.x.ai/v1` and works with the OpenAI SDK. Here is a minimal sketch. Check the current docs for exact field placement before you run it.

```python
from openai import OpenAI

client = OpenAI(base_url="https://api.x.ai/v1", api_key="XAI_API_KEY")

resp = client.responses.create(
    model="grok-4.7",
    input="List recurring 'X vs Y' and 'best X for Y' questions "
          "about invoicing tools. Quote each post and link it.",
    tools=[{"type": "x_search", "from_date": "2026-08-01", "to_date": "2026-10-31"}],
)
```

Budget for it. Since 21 September 2026, `x_search` costs $5 per 1K posts fetched plus $10 per 1K profiles. `web_search` is $5 per 1K calls, and grok-4.7 tokens cost $2 input and $6 output per 1M ([xAI pricing](https://docs.x.ai/developers/pricing)). The pricing page warns that costs "scale with query complexity," because the model decides how many tools to call. Set `max_turns`.

### Days 8 to 10: check the pattern against real volume

Grok cannot give you search volume. No xAI documentation mentions keyword data. Use Keyword Planner, Ahrefs or Semrush. Low-spend Google Ads accounts see Keyword Planner volumes as broad ranges like "1K-10K", according to [Search Engine Land](https://searchengineland.com/google-officially-throttling-keyword-planner-data-low-spending-adwords-accounts-255795). That report is old, but it is the best documentation available.

The 512-page case learned this the hard way. Its integration combo pages underperformed, so that team now requires search volume for both tools before it builds a combo page.

### Days 11 to 18: build and verify the dataset

Grok's code execution can dedupe and validate rows. It should not supply the facts in those rows. In the [Tow Center's February 2025 test](https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php), Grok 3 answered 94% of queries incorrectly, and 154 of its citations led to error pages. That was an older model. A [2025 arXiv study](https://arxiv.org/pdf/2505.18059) of bibliographic references put Grok first of 8 chatbots, but only 60% of its references were fully correct. I found no 2026 hallucination rate for Grok. Check every value against its primary source.

### Days 19 to 22: How to implement programmatic SEO

Backlinko suggests tooling by page count:

| Pages | Stack |
|---|---|
| 1 to 100 | Sheets plus WordPress or Webflow (WP All Import) |
| 100 to 1,000 | Airtable plus Webflow CMS or Make, or Hugo |
| 1,000+ | Next.js and Postgres with CI/CD |

For 50 to 200 pages, the first or second tier is enough. Backlinko says to build 3 to 5 test pages by hand first. Put all pages in one folder and give each template its own sitemap, so you can track indexing per template. Both points come from the 512-page case. More on choosing tools is in [programmatic SEO tools by stage and scale](https://maciejwiatr.com/blog/programmatic-seo-tools-by-stage-and-scale).

Don't chase word counts. Backlinko recommends 500 to 1,000+ words, yet the [Tool Pantry postmortem](https://dev.to/sebestian/302-pages-in-7-days-29-impressions-an-honest-programmatic-seo-postmortem-4372) found word count didn't predict indexing.

### Days 23 to 30: launch in batches

Publish the first batch, keep everything else on noindex, and put only indexable pages in the sitemap.

## How many programmatic pages should a new domain launch first?

Google gives no number. Its [crawl budget guide](https://developers.google.com/search/docs/crawling-indexing/large-site-managing-crawl-budget) says "every site starts with the same default, conservative crawl capacity limit." Practitioners point small:

| Source | Advice or result |
|---|---|
| Ahrefs | Test on about 10 rows |
| [Job board comparison](https://hackernoon.com/i-launched-two-similar-job-boards-google-indexed-one-and-rejected-the-other) | 10 to 20 indexable pages, rest noindex |
| 512-page case | 25 to 30 per week; one 80-page week cut that batch's indexing to 41% |
| Tool Pantry | 302 pages in about 7 days; only 7 ever shown |

In the job board comparison, a board that launched with many long-tail pages fell to 2 indexed pages and stayed there for about 2 months. The author's explanation for that is a theory and hasn't been confirmed.

My recommendation: build 50 to 200 pages, but make only 25 to 30 indexable in week one. Add the next batch once the earlier pages earn impressions.

## What day 30 actually tells you

It tells you about indexation, not rankings. Google says some changes "could take several months" ([SEO Starter Guide](https://developers.google.com/search/docs/fundamentals/seo-starter-guide)). Ahrefs suggests judging after about 3 months. Tool Pantry expects indexing to pick up in weeks 3 to 8.

On day 30, open the [page indexing report](https://support.google.com/webmasters/answer/7440203) and read the two statuses that matter:

- **Discovered, currently not indexed**: Google found the URL but hasn't crawled it yet.
- **Crawled, currently not indexed**: the page "may or may not be indexed in the future." Don't resubmit.

If most of the batch sits in "Crawled", stop adding pages and improve the ones you have. Some practitioners read this status as a domain trust signal. That is their inference, and Google's documentation doesn't say it. More in the [indexing hub](https://maciejwiatr.com/programmatic-seo/indexing).

## Which parts of a programmatic SEO build should not be handed to an AI model?

- **Data and facts.** Google's [AI content guidance](https://developers.google.com/search/docs/fundamentals/using-gen-ai-content) says "it is critical to manually factcheck and review all AI-generated content."
- **Body copy at scale.** The scaled content abuse policy names "using generative AI tools… to generate many pages without adding value for users."
- **Titles, meta descriptions, structured data and alt text** unless a person reviews them. Google's review requirement covers these too.
- **Choosing the pattern.** Grok surfaces candidate patterns. Volume tools decide which ones you build.
- **Rewriting scraped content.** The policy lists synonymizing, translating and "stitching or combining content" as examples.

## Can I do SEO myself?

Yes. Google's starter guide is written for site owners and says "There are no secrets here." It suggests hiring a professional only if you lack the time or interest. **Can a beginner or self-taught person do it?** Yes, on the same basis. **How difficult is it?** The concepts are learnable. The hard part is patience, because results can take months. Programmatic SEO adds technical work on top: structured data, internal links, sitemaps and keeping the data up to date, as [Semrush](https://www.semrush.com/blog/programmatic-seo/) notes.

**What are the 5 best SEO tools?** For this plan: Google Search Console, Keyword Planner, Ahrefs or Semrush, a spreadsheet or Airtable, and the Grok API for research. That is my pick, not a ranking. None of the sources I used covers the 80/20 rule, the "3 C's", SEO job demand or salaries, so I won't guess at them.

## The short version

Use Grok to find questions people actually ask, then confirm volume with a real keyword tool. Verify every data point yourself and hand-test 3 to 5 pages. Launch about 25 indexable pages and read the indexing report on day 30. Expect rankings in month 3, not week 4. If you have no data of your own, wait.