Cut your SEO budget because of AI? Here is how to recover.

It feels like decisive action. Vendor projections say search is being replaced, so the budget moves from SEO to "AI optimisation." In our 2026 report, it is the most expensive mistake we have observed.

Why it costs you

Brands that moved half or more of their search budget lost organic traffic and did not gain proportional AI visibility. The reason is simple. AI-sourced traffic is high intent, but for most businesses it is small in absolute share.

The two are also not separate channels. BrightEdge found that 54.5 per cent of pages cited in Google's AI Overviews were already ranking in the organic top ten. Cut the SEO and you weaken the very thing AI answers draw on.

How to fix it

1. Measure what the cut cost. Compare organic visits and enquiries for the three months before and after the change.
2. Restore the foundations first: technical health, the pages that used to bring in enquiries, and up-to-date dates on your key content.
3. Treat AI visibility as an extension of that work, not a replacement for it. Named authors, complete schema and well-sourced content help in both places.
4. Set a baseline. Check which AI assistants name you today, so your next budget decision has a number behind it.
5. Plan in years, not weeks. Authority compounds over two to three years.

The honest position

AI visibility adds to your search investment. It does not replace it. The brands growing their AI citation share did it by extending their existing programmes, not abandoning them. If someone is pitching you a separate AI package on top of your SEO, ask what they will do differently. Often the answer is not much.

In the report: Mistake Eleven (Chapter 11, page 51). BrightEdge data is referenced in Chapter 1.

Publishing more content with AI? Here is how to keep it from backfiring.

More pages used to mean more chances to be found. AI made more pages cheap. That bet has not paid off.

Why it costs you

The casualties are public. HouseFresh, Retro Dodo, Travel Lemming and others lost between 60 and 95 per cent of their traffic in the affected period, and recovery has been slow and partial.

Google's quality systems now treat content produced at scale as a quality signal, whoever or whatever wrote it. Its rater guidelines reserve the lowest rating for content that lacks human oversight and review. The rule is not anti-AI. It is anti-unreviewed.

There is a factual risk too. A 2026 audit found that 3 to 13 per cent of the source links produced by commercial AI tools were invented.

How to fix it

1. Pause volume. Stop publishing to a schedule set by output targets.
2. Review what is already live. Keep what is accurate and useful, and improve what is worth saving. Get advice before removing anything that still brings in visitors.
3. Use AI for research, structure and first drafts, not for the finished page.
4. Put a named expert on every piece. They review, correct and sign off, and their name goes on it.
5. Show publication and update dates, and disclose AI use where it has materially contributed.
6. Add what AI cannot: your own data, examples from your own work, and sources for your claims.

What works instead

Fewer, better pages. In the Princeton GEO study, adding citations, statistics and quotes from credible sources lifted AI citation visibility by 30 to 40 per cent on average. That is an editorial investment, not a volume one.

In the report: Mistake Six (page 50), Chapter 10 (page 47) and Chapter 12 (page 53).

No author names on your content? Here is the fix.

"By the team," "Admin," or no name at all. It is the default on most business websites, and it is quietly costing you.

Why it matters

AI answers increasingly favour content that ties its claims to an identifiable expert. An anonymous or generic byline gives a weaker signal than a named person with verifiable credentials.

Google's quality guidelines now treat trust as the most important of their four pillars. A real name behind the advice is one of the clearest trust signals you can offer.

How to fix it

1. Choose three to five authors. They should be real people in your business with genuine expertise, such as the founder, a senior practitioner or a specialist.
2. Give each one a profile page. Include their role, experience, credentials and photo, with links to LinkedIn and any professional registry.
3. Add Person schema to those pages. This is code that tells search engines and AI tools who the person is and links to their other profiles. It is a small job for a developer.
4. Add names to your most important pages first. Start with the ten pages that bring in the most enquiries, and show publication and update dates on each.
5. Build their presence beyond your website. Industry articles, podcasts, speaking and an active LinkedIn profile all add weight.
6. Publish an editorial standards page. Explain how your content is written and reviewed, and how AI is used.

Do the named authors have to write everything?

No. They must review it, correct it and be willing to stand behind it.

How long does it take?

Setting it up takes a few weeks. A 2025 analysis by Schema App found that brands that added article and author schema saw measurable gains in AI citations within two quarters.

In the report: Mistake Ten (page 51), Chapter 7 (page 34) and Chapter 9 (page 45).

Only watching clicks and traffic? Here is what to measure alongside them.

Clicks used to be a reliable measure of being found. They are not any more.

Why clicks alone mislead you

Pew Research looked at 68,879 real searches in March 2025. When an AI summary appeared, people clicked through to a website 8 per cent of the time, against 15 per cent without one.

That means your business can be named more often while your clicks fall. If clicks are the only number you watch, that looks like decline, and the natural response is to cut the very work that is producing results.

What to measure alongside clicks

• Citation share: how often your website is cited in AI answers to the questions that matter in your category.
• Brand mention rate: how often your business is named, with or without a link.
• Sentiment: whether what the assistants say about you is accurate and favourable.
• Share of voice: how often you are named compared with your competitors.

How to start

1. Write down 20 to 30 questions a buyer asks before choosing a business like yours. Over time, the report recommends building this to your category's top hundred.
2. Run them in ChatGPT, Gemini, Google's AI Mode and Perplexity, and record who gets named. Each assistant cites differently, so check more than one.
3. Connect the AI Performance dashboard in Bing Webmaster Tools, and check that Google Search Console is collecting AI Overview data.
4. Repeat monthly, with the same questions in the same format.
5. When you are ready, move to a tracking tool. Profound, Otterly, Peec AI and others are now within reach of mid-sized budgets.

Keep watching enquiries and sales. They are still the numbers that pay the bills. These new measures tell you whether the work that leads to them is building.

In the report: Mistake Four (page 50), Chapter 1 (page 7) and the Strategic Checklist (page 64).

Not on Reddit, Wikipedia or YouTube? Here is where to start.

These are the platforms the AI engines already trust, yet most businesses have never treated them as part of their own presence.

Why it matters

YouTube accounts for nearly 30 per cent of Google AI Overview citations. Reddit accounts for over 11 per cent in some studies, and over 6 per cent of Perplexity citations. Wikipedia sits in the top three sources on every major engine. Reddit's licensing deals with Google and OpenAI have made its content a primary source for the largest AI tools.

Where to start

Wikidata first. Wikipedia has a high bar for inclusion, especially for B2B businesses. Wikidata, its structured sister database, has looser criteria, and a complete record is itself a citable reference. That record covers your category, founding date, founders, services and official links. We have seen mid-sized businesses gain AI visibility from this step alone.

YouTube, if you can explain things on camera. Answer the questions customers already ask you. Add transcripts, captions, chapters and clear titles. The assistants rely on that text to understand a video.

Reddit, slowly and genuinely. Communities reject promotion. Businesses that succeed contribute helpfully for months, at roughly nineteen useful posts for every mention of themselves. Anything that looks like marketing gets spotted within hours.

LinkedIn, fully. Keep a complete company page and active profiles for your key people.

What not to do

Do not try to shortcut Wikipedia, because inclusion depends on independent coverage. Do not post promotional content on Reddit. Do not upload videos without transcripts or descriptions.

Choose one or two

Pick the platforms that suit your audience. Consumer brands usually get more from Reddit. B2B businesses should usually start with Wikidata and LinkedIn.

In the report: Mistake Seven (page 51), Chapter 7 (pages 33–34) and Chapter 9 (page 45).

Added schema once and left it? Here is how to finish the job.

Schema is code on your pages that tells search engines and AI tools exactly what things are: your business, your people, your products and your articles. Many websites have some. Few have it complete.

Two mistakes in one

Mistake Five is treating AI visibility as a formatting job. That means adding FAQ markup, bullet lists and contents boxes without improving the content underneath. The results are volatile.

Mistake Nine is partial schema. Examples include articles with no author, products with no price, and FAQ markup with no questions defined. It leaves the most consequential technical lever half-used.

Why it matters

Google says schema is not a direct ranking factor. It is, however, a direct signal AI tools use to ground their answers, and the gap between full and partial schema shows up in how often brands get cited.

How to fix it

1. List your page types, such as homepage, service or product pages, articles, author pages and location pages.
2. Match each page type to its schema type: Organization, Product, Article, Person, FAQPage or LocalBusiness.
3. Complete the important fields. Organization should link to your LinkedIn, Wikidata and industry registry profiles. Articles need an author and dates. Products need pricing and availability.
4. Test every template with a validator, such as the Schema.org validator or Google's Rich Results Test, and fix the errors.
5. Then improve the content itself. In the Princeton GEO study, the real lift came from citing sources, adding statistics and quoting credible experts, at 30 to 40 per cent on average. Markup describes your content. It cannot make it more credible.

How long it takes

A full schema audit, with each content type mapped, is roughly a one-week project. It is one of the few technical fixes that keeps paying off.

In the report: Mistakes Five and Nine (pages 50–51), Chapter 8 (page 37) and Chapter 9 (page 44).