SEO strategies for startups: keyword mapping, AEO and GEO

You launched three months ago, the ad budget is thin, and someone has told you that the right ten pages will bring steady traffic without paying for clicks. Maybe. Nobody can promise it. Google says that even a page that meets every requirement may not be crawled, indexed or shown, and that some changes take effect in hours while others take months.^[1]

What you can do is the work that Google and Bing document, in a sensible order, and measure it with the tools Google and Bing provide. This guide covers that work for a young site: how to build a keyword map, what "AEO" and "GEO" mean and what they do not, the on-page and technical basics, and links. Where the old advice in this post turned out to be unsupported, we removed it, and the last section says what went.

Why organic search suits a small budget, and what it costs instead

Paid clicks stop when the budget stops. Organic visibility does not switch off the same way, but it is not free: it costs writing, editing and technical care, and it arrives slowly. Google suggests waiting a few weeks before judging whether a change helped, and notes that not every change will show an effect.^[1] If you need leads this month, search is the wrong lever. If you will still be selling in a year, a site that is easy to crawl and clear about what it offers is an asset that keeps working.

We do not quote a return on SEO here. Any figure for "SEO ROI" depends on your market and your site, and the ones in circulation are not Google's.

Keyword research and mapping

A keyword map is a spreadsheet that assigns each search phrase you care about to one page. Its job is to stop two of your pages competing for the same query and to show where you have no page at all.

Start with seed phrases

List 10 to 20 phrases that describe what you sell, in the words a customer would use. If you sell a CRM, that might be "customer tracking", "sales pipeline software" and "client management for small teams". Add what customers asked you in sales calls and support tickets: those questions are real demand.

Add data from sources you can name

Three kinds of data are worth separating:

• Your own data. Google Search Console reports the queries your site already appears for, once you have verified the site.^[2]
• Google's planning tool. Keyword Planner gives ideas and estimates, but you need a Google Ads account, and Google says basic features such as keyword ideas require you to finish account setup with billing information.^[3]
• Third-party tools. These give volume and "difficulty" scores. Google says outside tools do not have its internal ranking data, cannot guarantee performance, and that any predictions are their own.^[4]

Treat "low competition" as a hypothesis. A difficulty number is a vendor's estimate. The check that counts is whether the pages that already rank for a phrase answer the same question you can answer, and whether you have something they lack, such as first-hand experience or a specific number from your own work. Gonzo does not supply keyword volumes, and our SEO audit page says why: real volumes need a licensed source.

Build the map

One row per page, with these columns:

The rows are an illustration with no search volumes, because we have none to offer. Two rules keep the map honest. First, one primary phrase per page, and no two pages with the same primary phrase and intent. Bing's webmaster team says near-identical pages split signals, blur which page should be shown, and can lead an AI system to pick the wrong version.^[5] Second, match intent. Someone searching "customer tracking spreadsheet" wants a how-to, not a product page, and a page that misreads the intent will struggle however well it is written.

AEO and GEO: what the terms mean

"Answer engine optimization" (AEO) is the work of getting your content used as a direct answer: in a featured snippet, a voice reply or an AI summary. "Generative engine optimization" (GEO) is the same idea aimed at AI search tools that write an answer and cite sources.

An earlier version of this post blurred GEO with geographic targeting. They are different things. The term GEO comes from a 2023 research paper by Aggarwal and colleagues, presented at KDD 2024, which proposed ways to raise a source's visibility in generative engine responses and reported gains of up to 40% in its own benchmark, with results that varied by subject area.^[6] That is a lab result on a test set, not a measurement of any live engine, and we do not repeat it as a promise.

What Google says about AI search

Google's guidance is plain, and it is the best available reference for anyone who sells "AEO services":

• To be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to appear in Search with a snippet. There are no additional technical requirements.^[2]
• There are no extra requirements to appear and no special optimizations needed. The existing SEO fundamentals apply.^[2]
• On "AEO" and "GEO", Google says that optimizing for generative AI search is optimizing for the search experience, and thus still SEO.^[7]
• You do not need to create new machine-readable files, AI text files or special markup. There is no special schema.org markup to add.^[2]
• You do not need to break content into tiny "chunks", rewrite it for AI systems, or chase every long-tail variation. There is "no ideal page length".^[7]
• Seeking inauthentic mentions on other sites is "not as helpful as it might seem".^[7]

It also warns against third-party tools that promise ranking success or claim to use "internal" Google metrics, and says to measure with Search Console.^[7] Search Console has a generative AI performance report for this.

What about llms.txt?

The llms.txt file is a proposal for a Markdown file at your site root that points language models to clean versions of your pages. The proposal's own site says thousands of sites publish one.^[8] It has no confirmed effect on rankings or citations. For Google the answer is explicit: Google Search does not use these files, and it says creating one will neither help nor harm your visibility in Google Search.^[7] The crawler pages of OpenAI, Anthropic and Perplexity that we read describe robots.txt as the control and do not describe llms.txt as a signal.^[9][10][11] If a tool you use reads one, adding it is cheap. Do not expect it to move a ranking.

Which crawlers to allow

This is the one AEO decision that is yours and has a documented answer. The major AI vendors run separate crawlers for different jobs, and each respects robots.txt:

OpenAI says each setting is independent: a site can allow OAI-SearchBot to appear in ChatGPT search while disallowing GPTBot to indicate its content should not be used for training.^[9] Anthropic describes its three bots in the same way, and notes that disabling Claude-SearchBot or Claude-User may reduce visibility in its search results.^[10] Perplexity describes PerplexityBot as surfacing sites in its search results.^[11] Which to allow depends on how you feel about training use. Blocking a search bot is a decision to be absent from that product's answers.

What Bing adds

Bing's webmaster blog introduced an AI Performance report in February 2026 that shows how often your pages are cited in Microsoft Copilot, AI summaries in Bing and select partners. It reports citations and the phrases used to retrieve them, and the post says it does not indicate ranking or the role a page played in an answer.^[12] Its advice reads like ordinary editing: clear headings, tables and FAQ sections, claims backed by evidence, and fresh, accurate content.

The local side of "GEO"

If customers find you by place, the relevant tools are old ones. Google's Business Profile guidelines say that a rented mailing address or virtual office is not eligible, and that service-area businesses should hide their address from customers.^[13] Keep the name, address and phone number consistent between your site and your profile, and put a real city or service area in the copy of any page that targets one. Do not build near-identical city pages in bulk: Google's spam policy on scaled content abuse covers many pages generated mainly to manipulate rankings.^[14]

On-page basics that still matter

Title and description. Google sets no maximum length for a title element or meta description, but it truncates results to fit the device.^[15][16] The old rules of 60 and 160 characters are habits, not limits, though a short title that puts the main phrase first is a good habit. Write for the person reading the result. Google may rewrite either one.

Headings. Use one clear heading for the page and sub-headings that split it into parts a reader can scan. Google says heading order does not matter for its systems, though it helps screen readers.^[1] Use secondary phrases only where they read naturally. Keyword stuffing is against Google's spam policies.^[1]

Internal links. Link related pages to each other with descriptive anchor text. This is how readers and Google find pages, and links are the main route by which Google discovers new ones.^[1]

Images. Compress them, give them short descriptive file names, and write alt text that says what the image shows. Google lists descriptive file names as a light clue and alt text as the most important metadata attribute.^[17]

Technical checks

• Allow crawling. Make sure robots.txt, and any content delivery network or host setting, does not block Googlebot or the search crawlers listed above. Google lists this first among the fundamentals for AI features too.^[2]
• Sitemap. A small site of about 500 pages or fewer that is well linked may not need one, but a new site with few external links might, and most content management systems generate one for you.^[18]
• Search Console. Verify the site and read the indexing and performance reports. Fix 404s on pages that used to matter by redirecting or restoring them.
• Page experience. Google says it highly recommends good Core Web Vitals and that this aligns with what its core ranking systems reward. Its targets are loading in 2.5 seconds or less, interaction response under 200 milliseconds and layout shift below 0.1.^[19] Those thresholds are a goal for users, not a switch that lifts a ranking.
• Structured data. It is optional for AI search, but still useful for rich results. Note what has changed: Google stopped showing FAQ rich results on May 7, 2026, and removed documentation for them in June 2026.^[20] The how-to rich result was retired in 2023, and Speakable markup is documented only for news publishers.^[20] Marking up an FAQ is harmless if the questions are visible on the page, but it will not earn an extra result in Google.

Links and mentions

Google finds pages mainly through links from pages it already knows, and links from other sites tend to appear over time as people find your work useful.^[1] Do not buy them. Google's link spam policy lists buying or selling links for ranking, excessive link exchanges, and links with optimized anchor text in guest posts or press releases distributed on other sites.^[14] Sponsored links are allowed when qualified with rel="nofollow" or rel="sponsored".^[14]

Guest posts are fine when a site's audience would genuinely want your piece and the link is incidental. Resource pages work the same way: ask only where your page belongs. Journalist-request services can also bring a mention. HARO, the best-known one, closed under its owner in December 2024 and was relaunched by Featured.com in April 2025, according to its Wikipedia entry.^[21] Whether a response earns a link is up to the publication, and nobody can guarantee one.

As for social posts, we found nothing in Google's documentation that describes likes or shares as a ranking input. Treat social as a way to be found and discussed, not a lever.

Where automation helps

Most of this list is judgment: who your customers are, what you know that others do not, and which question your page answers better than the one above it. Software helps with repetition. Gonzo's search and AI-answer agents read your own site, list what is in the way, and draft the fixes, and they only propose: a person on your side decides what gets applied. They do not give keyword volumes, control rankings or citations, or build links. Once people do find you, the next problem is what happens to the inquiry, and the Lead Response Engine is built for that step. If you are weighing an agency retainer against software, the comparison sets out where each wins. And if your site serves several countries, the next piece to read is our guide to international SEO mistakes.